Marine debris management is entering a new phase of data-driven applications. Converting years of accumulated coastal imagery into actionable tools for surveying, identification, and monitoring has emerged as a critical challenge in the digitalization of ocean governance.Under the guidance of Taiwan's Ocean Affairs Council (OAC), the National Academy of Marine Research (NAMR) is hosting the "2026 International Marine Debris Image Recognition AI Challenge." Featuring a dataset of over 20,000 real-world marine debris images, the competition invites AI, data science, computer vision, and marine science teams from Taiwan and abroad to participate.The competition is supported by Amazon Web Services (AWS) as the AI technology partner, with model evaluation and competition operations managed through the Industrial Technology Research Institute's (ITRI) AIdea AI Co-Creation Platform. Registration is now open.NAMR sets the challenge: bringing AI to the frontlines of ocean governanceMarine debris has long been a fundamental issue in coastal environmental governance - and one of the most difficult to address in the field. Coastal debris is diverse in type, scattered in distribution, and frequently degraded by sun exposure, seawater erosion, sand burial, and physical damage, making manual surveys and image interpretation highly labor- and time-intensive.To accelerate digital transformation, NAMR has established MDImageNet, an AI-Ready marine debris image dataset covering the ICC19+1, NAMR26+1, and NAMR33+1 marine debris category schemes. The competition draws on NAMR's existing marine debris image dataset, comprising over 20,000 images annotated with YOLO-format bounding boxes. The dataset covers common coastal waste categories including plastic litter, fishing-related debris, and other anthropogenic waste.A "post-mapping" strategy is adopted: participants first train models using the original class labels provided, then map predictions into 20 official recognition categories during inference, with final scoring based on 19+1 primary marine debris target classes.The competition design confronts teams with the real-world constraints of field data - cluttered backgrounds, diverse object classes, and significant appearance variations. By requiring quantitatively evaluated object detection models, the challenge goes beyond open data sharing: it validates whether AI can be transformed into deployable tools built upon existing survey infrastructure.The competition comprises preliminary and final rounds, evaluated primarily on mean Average Precision (mAP) at an IoU threshold of 0.5. A "Best Lightweight Optimization Award" is also offered, using NetScore to balance detection accuracy, model size, and computational complexity-encouraging models suitable for practical deployment in coastal patrols, UAV-based image analysis, and long-term environmental monitoring.AWS cloud resources and AIdea platform power hands-on AI developmentThe competition integrates resources from AWS and ITRI's AIdea AI Co-Creation Platform to support teams throughout model development, training, testing, and evaluation. AWS provides the cloud development environment, offering Amazon SageMaker AI and computing resources for machine learning workflows. Technical workshops are also scheduled to familiarize participants with cloud-based AI development tools and model training pipelines.ITRI's AIdea Platform handles competition execution, dataset support, and automated scoring. The standardized cloud environment and evaluation framework ensures all teams operate under identical conditions, maintaining fairness and reproducibility.The competition is open to high school students, college students, and professionals, with teams of two to five members. Cross-institutional and interdisciplinary collaboration is encouraged, combining expertise in AI, computer vision, data science, and marine science to produce evaluated model outcomes for marine debris recognition.Registration for the "2026 International Marine Debris Image Recognition AI Challenge" is open until August 10. The total prize pool is NT$300,000, with two additional Best Lightweight Optimization Awards.NAMR aims to bring together the AI community, research institutions, academia, and industry to convert marine debris imagery into deployable environmental monitoring models-building Taiwan's AI application experience and marine data foundation for ocean governance.For competition details and registration, visit the official website (link).

Communication – for whatever purpose, be it trying to persuade the electorate to vote for you, to convince your audience to buy whatever product you're selling, to manage client relationships, or simply to become famous online through creating engagement – is, at its heart, an exercise in storytelling. In today's visual-first world plagued with short attention spans and what can seem like an infinite number of e-mails and ads diverting our concentration, being able to fashion compelling video-based stories is increasingly crucial to successful communication. Attempts to automate communication can also seem generic and devoid of empathy and personalisation, consigning them to failure.EQX Lab and its product, EQX Flow, aim to use artificial intelligence to enable the masses to successfully grab others' attention. By inputting a very simple text prompt, users will be able to generate videos that have the potential to go viral. Corporate sales teams will also be able to automate the creation of personalised, cinema-quality videos for client retention purposes.EQX Lab is the brainchild of Benjamin Cheung. He studied computer animation at the Savannah College of Art and Design in the state of Georgia and then moved to Hollywood, where he worked for many of the major animation studios – Square, Disney, DreamWorks, Sony, and Lucasfilm – for roughly 15 years. He worked on Square's Final Fantasy: The Spirits Within, which came out in 2001 and was the world's first photorealistic computer-animated film (where the animated images appear as real-to-life as possible); and when he was at Sony Imageworks, on 2007's Beowulf.But Cheung always knew that he wanted to return to Asia, and so he moved to Taiwan in 2010 to work for an animation company that counted some of his former employers as major clients. He was then poached by a cloud-computing company in China whose president at the time was Taiwanese. Cheung stayed with this company, which specialised in rendering (using a computer to generate [usually 3-D] images) for almost seven years. During this time, he grew the company’s staff count by roughly seven times and its revenue by a factor of 30.Cheung had been the technical director for most of his projects in Hollywood – he was constantly using new tools and had to develop their back-end integration himself. This gave him enough confidence in his technical abilities to launch EQX Lab. He realised that even though there were already plenty of AI video-generation models, they all required a detailed prompt and only generated a somewhat passable video.Instead, what if you could make a professional-quality video with just a simple click, or a single word, or a short sentence? That's what EQX Flow aims to do.The tool is powered by the company's NEXES Engine, an AI pipeline that, from a single piece of user input, generates prompts, writes a script, composes a storyboard, and finally outputs a video. Chain-of-thought reasoning is used to come up with an emotional arc and narrative to the story. The system handles everything internally, so that the user doesn't need any technical knowledge to produce a cinematic result. Moreover, the engine creates five stories and, again using AI, measures how gripping they will be, choosing the best one to increase the chances of the chosen story going viral. (The engine pays particular attention to optimising the first three seconds of the clip, as these are the seconds that decide whether users continue watching or not.)In the current prototype, users input a story idea (something as simple as "play tennis"), and then specify from a list of options the gender of the protagonist; the type of video (e.g., a sports/fitness video); its niche (e.g., a glow-up video); the length of the clip; the type of person that the creator himself/herself is (e.g., a no-nonsense expert); and the 'look' of the clip (e.g., commercial realism, cinematic realism, cyberpunk, vintage film, or anime). Cheung intends to add a 'director's cut' option in a later incarnation of the story engine, where users will be able to emulate, for example, Christopher Nolan's aesthetic.EQX Flow will have two modes: fast and professional. The latter will be more sophisticated and, as a result, also more expensive (in his current prototype, it is about 2.5 times as expensive as the fast model). A tennis ball hitting a player in the face would look passably realistic in the professional-mode video; the fast mode would instead produce stylised visuals in a fraction of the time. Cheung says that directors often prefer the fast mode because it can produce unexpected scenes that they consider to be spontaneous.The tool is also modular: users will be able to stop after each stage (e.g., script generation), tweak things, and then move onto the next stage. Directors who already have a script and storyboard might not need the story engine at all – instead, they will only use EQX Flow’s back end to generate the shots and make the actual video. EQX Flow currently uses Google's generative AI models. This does not allow for the creation of violent, graphic, or adult material, a limitation that is in line with EQX Flow's business model and vision. They are also developing a model-agnostic pipeline, and plan to integrate additional leading AI video models to expand their creative possibilities and to reduce platform dependency.The company's Face-Lock technology can also incorporate photos that users upload of themselves or their desired characters into the videos.If non-specialists can now produce videos, should directors be worried about the future of their profession? Cheung doesn't think so. He says that average directors who don't learn how to use AI tools and don't improve the quality of their work might very well be replaced – but a good director can't be replaced by AI because directing is an art form and good films are unique products that cannot simply be reassembled from pre-existing films. Although Cheung is appreciative of the award, he is aware that he needs more than just the money it brings to fully launch EQX Flow. He is drawing on his network to expand his core technical team, growing his corporate presence in Taiwan, and actively seeking venture capital investors. In particular, he would like to raise three years of funding for research and development, which he will carry out in Taiwan. Cheung says that Taiwan provides a stable environment for his company, with a robust legal framework for intellectual property rights, top-tier engineering talent, and unfettered access to services such as Google.But despite its physical presence in Taiwan, Cheung is focussed on the North American market. Hollywood has the most advanced technology and the most resources for storytelling; Google works best in English; and the story engine has also been developed in English.The Best AI Awards celebrate global excellence in artificial intelligence and IC design, welcoming submissions from innovative companies and brilliant student teams. Following the success of the 2026 edition—advised by the MOEA, organized by DoIT, and executed by TCA—the prestigious competition is officially transitioning into an annual tradition.Offering substantial grand prizes and unmatched industry exposure, the countdown to Best AI Awards 2027 has already begun. Details on the next submission cycle, prize tiers, and eligibility rules will be released soon. Connect with us on LinkedIn for the latest official updates and application alerts.

"Integrated circuit" is a fancy term for the chips that control the electronics that make our modern life possible. They are composed of hundreds of thousands or even millions of tiny components, such as transistors, capacitors, and resistors. These components are so tightly linked together on a tiny piece of silicon that, for all intents and purposes, the chip is indivisible – you can't separate it into its individual components.Any mistake in a chip's circuitry could thus render the entire chip unusable. The process of transferring the as-yet theoretical chip design to the fabrication plant (also known as a foundry or chip fab) to produce a photomask (the template that will later be used to create chips en masse) is known as the tape-out, and costs in the tens of millions of dollars.So chip designers have a strong financial incentive to get their designs right on paper before any silicon is involved.But research conducted for Siemens in 2024 showed that only 14 per cent of chips experienced what is called first-silicon success – when the first tape-out results in a photomask and a physical chip that reliably works. This research also noted that this was the lowest success rate over the past two decades. One reason for that might be that many companies are now bringing chip design in-house instead of contracting out to experienced, specialised chip designers.That's where Vincent Bligny's company, Aniah, steps in.Based in Grenoble in the French Alps, Aniah uses artificial intelligence to conduct electrical rule checks (ERCs) for integrated circuits. These checks examine the electrical connections between different components and verify that the chip design can be made, will work, and will continue to work over time. For example, ERCs look for short circuits, when current flows with very little resistance between two components that it was not supposed to flow between; and they look for open circuits, the opposite of a short circuit, where the circuit is not a closed loop and current therefore cannot flow.ERCs also look for large voltage drops (often because of high resistance in the circuit), leading to overheating, noise, and reduced efficiency; and they look for voltage violations, where two components operating at different voltages are connected, potentially damaging the chip (level shifters are necessary to get around this problem). Traditional checkers, such as the industry-standard simulator SPICE (Simulation Program with Integrated Circuit Emphasis), examine how the chip might behave under different voltage, current, or temperature conditions. However, they are unable to sample every potential configuration of the circuit’s components and might miss corner cases (the rare cases where the parameters are at the limits of their expected range). Aniah's OneCheck, on the other hand, covers all of these potential configurations. If there are N components that can take on either high or law values (e.g., N power domains that can be switched on or off), OneCheck will examine each of the 2N potential states to make sure that the chip will still run correctly. It will go over all possibilities, catching problems such as conditional high-impedance (floating) nodes and electrical-overstress violations in chips that mix high- and low-voltage domains. Unlike traditional checkers, OneCheck doesn't depend on luck: it is a static, simulation-free method. Moreover, traditional checkers might find the first instance of a particular type of violation and immediately report back that this violation is present. The designer might fix this one violation, but many other instances of this violation could still remain. OneCheck, on the other hand, will look for all instances of the violation, so that the designer can fix them all at once and not have to rerun the checker.Not only is OneCheck more comprehensive than traditional electrical rule checkers but it is also faster, giving results in minutes instead of days. There is a trade-off, of course: OneCheck finds more false positives than traditional checkers. In other words, it flags more violations that really aren’t violations at all. However, this is something that most chip designers can accept, especially because of another product that Aniah released earlier this year, Amigo.Amigo is an AI agent that goes hand-in-hand with OneCheck. OneCheck flags potential violations and groups them into roughly fifty clusters, organised by their root cause. Amigo then provides users with an explanation of these violations, and takes advantage of natural language processing – users can simply enter "Explain this violation" – to allow users to query and thus debug these violations. But Amigo's support for chip designers goes further than that. It also provides concrete suggestions for how to fix the violations that it has identified, which the chip designer can then validate. It also tells chip designers which violations are the 'lowest-hanging fruit,' whose correction would lead to the greatest increase in the likelihood that the chip will be ready for tape-out. In fact, Amigo quantifies for users just how ready a design is for tape-out, and can quickly adjust this valuation when changes are made (instead of running the entire checking process from scratch).OneCheck and Amigo are therefore a newer, AI-enabled system for electronic design automation (EDA), and Aniah has ambitious plans for developing this EDA technology over the next two years.  They intend to implement parallel fixing by the end of 2026, where the checker can make corrections as it checks. This is a prelude to delivering results in just seconds by next year.Aniah next wants to go beyond OneCheck and simply electrical rule checking to full circuit sign-off – in other words, it wants to be able to verify that physical design constraints are respected. In the long term, Aniah wants to enable continuous sign-off, whereby all electrical rules and physical constraints are constantly satisfied. Currently, sign-off is a hectic scramble at the end of the design process, where bugs are easy to overlook.Their bronze award at the 2026 Best AI Awards came with a prize of NTD 500,000 (USD ~$16,000). The money itself won't help Aniah a lot – they've already raised more than USD $11 million in funding and expect to open a new funding round soon – but the award will open doors with Taiwanese companies and investors, said Allen Chen, Aniah's director of applications engineering. Chen, who is based in Taipei, added that Bligny, Aniah's CEO, has long held Taiwan's technical prowess in semiconductor manufacturing – its world-leading foundries and design houses – in high esteem. This is why Bligny plans on expanding their Taiwanese team, said Chen, who joined the company three years ago and is currently Aniah’s only employee on the island.Aniah currently has one Taiwanese customer, Novatek, but also supports the 200-strong Taiwanese team of Nvidia, an American company. These clients are loyal to Aniah because OneCheck and Amigo offer the fastest running time and the smallest number of false positive alerts, and because Aniah is quickly adopting AI to improve its service.The Best AI Awards celebrate global excellence in artificial intelligence and IC design, welcoming submissions from innovative companies and brilliant student teams. Following the success of the 2026 edition—advised by the MOEA, organized by DoIT, and executed by TCA—the prestigious competition is officially transitioning into an annual tradition.Offering substantial grand prizes and unmatched industry exposure, the countdown to Best AI Awards 2027 has already begun. Details on the next submission cycle, prize tiers, and eligibility rules will be released soon. Connect with us on LinkedIn for the latest official updates and application alerts.

Artificial intelligence is just an interesting theoretical problem for many scientists and engineers, but it is at its most useful when it directly responds to the needs of its users.That's why Amity Solutions, a component company of Thailand-based Amity Group, developed Eko Agentic, a data analyst for store managers. They were long-standing consultants to one of the biggest retail chains active in Thailand and Malaysia with thousands of stores in the region. Executives at this retailer told Amity Solutions that they had tried to use various AI tools to improve the efficiency of their store management, but that these tools were not adequate for their needs.Store managers, stock replenishers, and other frontline workers on the store floor have to handle numerous disconnected tasks on a daily basis. They might use dashboards to monitor various store performance metrics, but synthesising the disparate information into business decisions can be complicated, with store managers resorting to past experience and guesses. Inexperienced store managers, in particular, might be unable to respond effectively to new situations or to best implement requests from headquarters.What if AI could take over the data analysis from store managers? Amity Solutions developed Eko Agentic to do just this: It is trained with data on how the top-performing store managers across the retailer's large network would respond to various business situations, and then rolled out across other stores, taking into account each store’s particular characteristics. The goal is to reduce extra, unsellable stock; to avoid empty shelves; and to better time and set up promotions. This way, the retailer tries to make all stores as efficient as those run by the best store managers.In its first iteration of Eko Agentic, Amity Solutions identified those stores that consistently outperformed the average, both through looking at store performance metrics and by talking to headquarters. Positive outliers were also identified in different environments – for example, the best inner-city markets (which tend to be smaller) and the best rural hypermarkets (which tend to be larger) – in order to get the widest possible range of data.Amity Solutions then sent teams to perform interviews at each of these stores, asking frontline workers to explain how they would think through various situations. What would they do if sales dropped by 5 per cent year-on-year? Perhaps the store manager would first check the basket size, then check the average value of each item in the basket, and then check for the use of special promotions.AI – and in particular, a technique developed by Amity Solutions called reflective optimisation via automated debugging (ROAD) – then structured these interviews into decision trees that visualised the store managers' train of thought. Most optimisation methods so far rely on large data sets for testing and calibration, but these interviews with store managers at Lotus's produced a smaller data set, something that ROAD's algorithm could work with. This was especially important in the Thai context because most large language models are trained on Western datasets, but differences in culture and the business environment between the West and Thailand (e.g., in the availability of parking lots) meant that other models, trained on larger data sets, weren't necessarily immediately applicable.The model was then applied to each individual store, generating a strategy that had been optimised for each one. Reinforcement learning (a paradigm within machine learning that seeks to optimise the impact of an agent's actions based on continual feedback to that agent from those impacts) is then used to further optimise store managers' strategies.Despite this rather simplistic description (and its correspondingly smaller size), Eko Agentic has been remarkably effective in data analytics. It is cheaper than many other AI tools (such as Claude and ChatGPT), and outperforms other state-of-the-art LLM and AI data analysis agents in an industry-standard set of real-world problems, the Data Agent Benchmark for Multi-step Reasoning (DABStep). It achieved 41 per cent accuracy in resolving DABStep tasks – the highest amongst all such agents – whilst its nearest competitor, Microsoft, only achieved 32 per cent accuracy; Anthropic's, OpenAI's, and Google's systems lagged even further back.Eko Agentic is also now able to outperform human analysts working at the Thai retailer. A blind test was conducted, wherein both Eko Agentic and a human analyst performed an analysis on various real business problems. Store managers then select the better of the two responses, without knowing who composed each one. The first versions of Eko Agentic still performed below a human analyst, but the latest version – the fifth – gives, on average, suggestions that are favoured over those from a human analyst. There are only a few supermarket chains in Thailand, and Amity Solutions is, of course, unable to work with the competitors to the retailer it currently works with. However, their methodology is applicable to other retail applications – and in fact, Amity Solutions is currently using Eko Agentic to help a telecommunications giant in Thailand manage its mobile phone shops. Amity Solutions is also looking for opportunities to apply Eko Agentic to supermarket chains in other Southeast Asian countries.A potential limitation with basing decisions on what the best store managers would do is that one might be limited to – and thus not be able to improve on – how well the best store managers do. In other words, you can interpolate performance but it is uncertain whether you can extrapolate to even superior strategies.Thus, as one of its next steps, Amity Solutions is creating a large behavioural model (LBM) that serves as a stand-in for customers. It is a digital twin that simulates customer behaviour, and models that respond to this LBM can potentially outperform the current best store managers.Amity Group, with offices in Thailand, Malaysia, Singapore, Australia, India, the United Kingdom, and the United States, employs 800 staff members over five companies in various realms of AI. Amity Solutions, the business unit that commissioned Eko Agentic through its long-standing collaboration with the aforementioned retailer, is based in Bangkok and employs 150 employees. However, it was Amity's AI Research and Application Center (ARAC), whose small team of just 15 staff members deploys generative AI solutions across all of Amity's daughter companies, that developed the technology behind Eko Agentic. Currently based in Thailand, they aim to stay at the forefront of developments in AI, says Touchapon Kraisingkorn, the chairman of ARAC and the executive director of Amity – and thus they have plans to expand ARAC to Singapore.Winning at the Best AI Awards is, says Kraisingkorn, validation that they are "one of the world-class labs that creates an effective product and solves real-world problems". The earnings from this award will help them jump-start hiring in Singapore. They are also open to opportunities for collaboration with Taiwanese companies in chips and robotics. The Best AI Awards celebrate global excellence in artificial intelligence and IC design, welcoming submissions from innovative companies and brilliant student teams. Following the success of the 2026 edition—advised by the MOEA, organized by DoIT, and executed by TCA—the prestigious competition is officially transitioning into an annual tradition.Offering substantial grand prizes and unmatched industry exposure, the countdown to Best AI Awards 2027 has already begun. Details on the next submission cycle, prize tiers, and eligibility rules will be released soon. Connect with us on LinkedIn for the latest official updates and application alerts.

Cardiovascular disease – diseases of the heart or the blood vessels that move blood to and from the heart – were responsible for an estimated 32 per cent of deaths worldwide, or just under 20 million deaths, in 2022, according to the World Health Organisation. Their takeaway? "It is important to detect cardiovascular disease as early as possible so that management with counselling and medicines can begin."Electrocardiograms (ECGs), a series of peaks that represent each heartbeat, depict the heart's electrical activity over time. An abnormal ECG (compared to one's baseline) can be indicative of arrhythmias, the medical term for irregular heartbeats that, in most cases, are not serious but can sometimes lead to strokes, heart attacks, and death. Patients can wear Holter monitors to continuously measure their ECG, but that entails placing a large number of electrodes (between three and eight, and up to twelve for greatest accuracy) on the skin, and wearing a piece of recording equipment around the neck or waist. Not only can this be inconvenient, but up to half of all patients reported some sort of skin irritation due to the electrodes.Smartwatches can also be used to measure an ECG, but only when both hands touch the device – in other words, they cannot be used for passive, continuous heart monitoring. That's where integrated circuits can help. Students at the Hanoi University of Science and Technology (HUST) are working on an integrated circuit that can continuously monitor the heart's electrical activity through the ear. Such devices are already commonplace, with many people wearing hearing aids or smart hearables for at least some of the day. There are three physical connection points – both ears and one earlobe – the minimum required to measure an ECG.The ECG collected between the ears is useful only insofar as it can then reconstruct what lead-1 ECG, which is the electrical activity as would be measured by electrodes placed on the right and left arms. The lead-1 ECG is also a standard ECG measurement that is used for diagnosis and monitoring arrhythmias. There are generally similarities between the ear ECG and the lead-1 ECG – in particular, the peaks (representing the heartbeats) appear at the same location – but they are clearer in the lead-1 ECG than in the ear ECG.So while the algorithm on the chip must be able to recover the shape of the original lead-1 ECG with as little noise as possible, it should not smooth over any possible signs of abnormality – in other words, it needs to be sensitive enough to detect arrhythmias when they are present.The system is relatively unobtrusive and runs on low power, but using the ear also presents disadvantages. The signal-to-noise ratio is low. In fact, simply shaking one's head or talking will introduce noise into the measurements. Furthermore, special data privacy concerns when collecting biological signals – each user has a unique ‘heartprint’ and some users might be particularly wary of sending such data to another machine – make it imperative that any analysis is done on the chip itself.The user himself can measure his own lead-1 ECG using his two fingers (as proxies for the right and left arms), and measure his ear ECG with the devices touching his ear. All of this is done with electrodes attached to a sensor (with a built-in analog-digital convertor, or ADG) developed by Texas Instruments. Two datasets were created to train the AI calibration algorithm for the conversion between the ear and lead-1 ECGs. Firstly, the team collected its own dataset, measuring the ECGs for 45 patients for 10 minutes each. A synthetic public dataset was also created by modifying an existing large, open dataset of ECGs, the PTB-XL. This dataset doesn't include ear ECGs, so the team added noise to existing lead-1 measurements (in an attempt to emulate the ear ECG) and tried to recover the original, non-noisy lead-1 ECG. The team, which calls itself EDABK Brain, was able to bring the algorithm's latency, or delay time between receiving the ear ECG and producing the lead-1 ECG, down to below 50 milliseconds. Such short times obviate the need to store data from the ear-ECG, for instance. At the same time, they maximised the utilisation of the processing element, meaning that they worked hard to make sure the chip was effective.After prototyping a field-programmable gate array (FPGA) with their IC design, the team trained it on the self-collected dataset. On the two most important metrics, the signal-to-noise ratio and the correlation with the true lead-1 ECG, EDABK Brain slightly outperformed state-of-the-art algorithms. It was edged out by another algorithm – but the HUST model used fewer than a quarter as many parameters as that algorithm did.For another comparison, EDABK Brain's circuit used less power and was more energy efficient than BioGAP, a leading biosensing platform that can measure ECGs as well as other electrical signals in the body. However, BioGAP's circuit has a lower latency time and a high throughput (measured in operations per second).The team behind this chip design, the EDABK Brain Team (the EDA stands for electronic design automation, and the BK stands for Bách khoa, which is in the Vietnamese name of their university,  consists of Phuong Linh Nguyen, a student who graduated from their university last year and is now studying for a master's degree at Télécom Paris (one of the most prestigious French grandes écoles and part of the Polytechnic Institute of Paris), and who spoke to DIGITIMES; and her two former classmates, Thanh Dat Do and Duc Tu Nguyen, both of whom are in their final year in the School of Electronics and Electrical Engineering at HUST; and their supervisor, Duc Minh Nguyen. Having the chance to participate in the Best AI Awards motivates Nguyen and her teammates. Being just students at the start of their scientific career, they were curious to know what industry professionals thought of their idea – does it have potential? Winning the bronze medal is confirmation that it indeed does have potential.Nguyen said that it was also a relatively rare opportunity for them to communicate their ideas in a non-academic setting. Their university was able to send three teams to the finals, which also entailed a trip to Taiwan and interactions with state-of-the-art AI and IC researchers.They will now focus on writing a paper – after all, they come from academia – and preparing patent applications. On the technical side, they want to reduce the number of bits in their resolution – in other words, see whether they can convert the analog signal to a digital one with a fewer number of bits and less accuracy – to reduce complexity and thereby power consumption. The team will also experiment with other electrodes.The Best AI Awards celebrate global excellence in artificial intelligence and IC design, welcoming submissions from innovative companies and brilliant student teams. Following the success of the 2026 edition—advised by the MOEA, organized by DoIT, and executed by TCA—the prestigious competition is officially transitioning into an annual tradition.Offering substantial grand prizes and unmatched industry exposure, the countdown to Best AI Awards 2027 has already begun. Details on the next submission cycle, prize tiers, and eligibility rules will be released soon. Connect with us on LinkedIn for the latest official updates and application alerts.

The Best AI awards were given out in two categories: artificial intelligence (AI) applications and integrated circuit (IC) design. AIYO was entered in the IC design category, but their concept – AI-powered acceleration of IC design – really straddles the two categories.Chip design is a long, arduous process. Today's chips contain billions of transistors, and designers not only have to design them to do what they want them to do, but have to ensure that the design satisfies a large number of rules and can actually be physically produced in a foundry. Since 2010, the number of transistors and gates has increased by nearly a hundredfold, but engineering productivity (as measured by the number of gates that a chip designer can design in a day) has increased by only about three times. This 29-times gap is set to widen even further over the next few years. In other words, it takes chip engineers more time than ever before to do their job.The problem lies not solely in the number of components on a chip but in the complexity in how they interact. The demand for custom silicon – ICs specially designed and optimised for specific applications or customers – and the sheer number of use cases now being designed for is growing, but it is getting harder and harder to find the engineering talent for custom silicon. Specialised silicon requires specialised talent.Tier-one design houses, like NVIDIA or MediaTek, can still find talent, but it's a different story for tier-two companies. Chip design is increasingly a bottleneck for technology companies to implement their AI-powered (and non-AI-powered) solutions.AIYO thinks that AI can help with that. It aims to reduce the gap between tier one and tier two companies, without completely replacing human design and the need for engineers. Engineers use natural language to provide their design specifications, including the constraints and requirements that the chip must satisfy. AIYO's AI agent then generates Verilog, which is a piece of code that describes the design of digital circuits.This Verilog must satisfy the specifications that the user set out. AIYO's agent can also optimise it along various metrics, such as for a lower power consumption, a higher performance (in other words, how fast the chip operates, as measured by its clock speed or data throughput), or a smaller area (PPA) are the most commonly invoked. Different chips prioritise these three factors (collectively known as PPA) differently, but AIYO can trade off, for example, a lower performance for a lower area and less power consumption.AIYO's agent then performs a closed-loop verification of the design to ensure that it is feasible. The agent reviews the error logs and iterates the design until the design passes all tests. Then – at least theoretically – the design is ready for tape-out, or actual production of the circuit at the foundry. Tape-out is an expensive process, costing in the millions of dollars, so it is essential that the finalised design performs as expected.AIYO uses RISC-V, an open-source instruction set architecture (ISAs) that has grown exponentially in popularity since its introduction in 2014; it has already been used in more than 20 billion cores. This avoids the need to pay a licensing fee to the more common (but not open-source) ISAs in use today, such as ARM or x86.The performance of IC-design agents can be measured using a standard set of test problems. Can the agent solve the problems (i.e., design a suitable IC) on the first pass? And can it solve a different set of IC design problems eventually, after however many iterations? AIYO performs at the head of the pack in both models, a little bit ahead of NVIDIA's VerilogCoder agent and far ahead of ChatGPT, DeepSeek, and Claude. One to two engineers are now required to design a chip, where three to five would have been needed before. Furthermore, it now takes these one or two engineers two months to design a chip (and they can test multiple designs simultaneously), whilst those three to five engineers in a traditional design house would have needed six months. This is an improvement in efficiency of roughly an order of magnitude.AIYO is still a work in progress, and humans are needed to check for any mistakes the engine might make – human engineers' jobs are safe, at least for now. But AIYO does makes it possible to compress the iteration cycle and for even tier-two chip design companies to realise their designs with limited human resources.One of AIYO's most pressing next steps is expanding its customer base – not only for commercial reasons, but also because this will expand their training data and thus improve their AI engine. AIYO currently has one customer who needs help designing custom FPGA (field-programmable gate array) integrated circuits. AI is, by default, a generalist, and its large language models can fail when faced with very specific use cases that it has yet to see. Helping this customer with its specific use cases can help the AI engine gain specialist skills, and the more customers that AIYO can obtain, the more versatile the tool will be.To this end, the team is expanding the set of design problems. A large set of open-source IC design problems already exists, but AIYO is also working on building their own set of synthetic design problems.The money from the Best AI Awards is great, says Ballard, but realistically, it is not even enough for one year's access to a top EDA (electronic design automation) tool – training AI models is expensive. He says that the biggest benefit from the awards is the people it has allowed them to meet, and in particular, the conversations they've been able to have. "They'll ask, 'Did you consider X, Y, and Z?' Sometimes yes, we have, but sometimes, we have an action item for the future."This also gives them an opportunity to enter talks with various venture capital investors and potential Taiwanese partners. AIYO is currently working with funding provided by the co-founders themselves, but they're hoping to find a Taiwanese venture capital investor in the next few months.Tang-Hung Po and Austin Ballard have been working on AIYO for only roughly a year. Po, originally from Taiwan and now based back in the country, obtained his master's in electrical engineering and computer science from the University of Michigan; he is the company's primary engineering lead. He brings more than 20 years of experience in SoCs (systems on a chip) and ASICs (application-specific integrated circuits) to AIYO, and was previously a director and a chief technical officer at other companies.Ballard, an American based in Seattle but with a Taiwanese mother, brings his experience scaling operations at Meta, Amazon and TikTok to now handle anything at AIYO not related to engineering,  The time difference allows them to collaborate during Ballard's evenings and Po's mornings, and their almost diametrically opposite locations, along with their different skill sets, facilitates engagement with all sorts of partners on both sides of the world. (As a side benefit, Ballard now has a business reason to visit Taiwan!)The Best AI Awards celebrate global excellence in artificial intelligence and IC design, welcoming submissions from innovative companies and brilliant student teams. Following the success of the 2026 edition—advised by the MOEA, organized by DoIT, and executed by TCA—the prestigious competition is officially transitioning into an annual tradition.Offering substantial grand prizes and unmatched industry exposure, the countdown to Best AI Awards 2027 has already begun. Details on the next submission cycle, prize tiers, and eligibility rules will be released soon. Connect with us on LinkedIn for the latest official updates and application alerts.

The Ministry of Economic Affairs officialized the successful conclusion of the second annual Best AI Awards, held at the Taipei Nangang Exhibition Center Hall 2. This year's event witnessed an extraordinary surge in international engagement, drawing 1,487 competing teams from 36 countries across the globe, which represents a near two-fold growth in foreign participation compared to the inaugural edition. Out of 253 finalist teams, 100 awards were presented to recognize exceptional breakthroughs in artificial intelligence technology and integration.Leading Innovators Secure Top Honors Across Key Industry VerticalsThe competition highlighted highly competitive solutions spanning healthcare, information and communications technology, manufacturing, and education. Eight prestigious gold medals were awarded to outstanding organizations and academic institutions, including Delta Electronics, Kneron, BrainNavi Biotechnology, ZenTech, National Cheng Kung University, and National Formosa University, alongside pioneering international teams from Thailand and Poland. These entries showcased the practical implementation of artificial intelligence, bridging advanced research with market-ready products.Minister of Economic Affairs Kung Ming-hsin. Credit: TCAMinister Highlights Shift From Hardware Dominance to Practical Industry DeploymentDuring the ceremony, Minister of Economic Affairs Kung Ming-hsin emphasized that while Taiwan maintains a critical global advantage in foundational AI hardware such as semiconductors and servers, the next vital phase involves translating this power into practical applications across all sectors. Under the framework of the government's new major AI infrastructure initiatives, the ministry has established over 50 trial production sites with automated AI capabilities and developed models covering 23 core industries. More than 1,000 consultants have been deployed across the nation to actively assist over 2,600 enterprises in integrating artificial intelligence into their daily operations.Emerging Tech Trends Shape the Future of Edge and Agentic IntelligenceThe second edition successfully guided industrial investments toward the frontier developments of Agentic AI and Edge AI, shifting artificial intelligence from a passive responsive tool to an autonomous partner capable of independent judgment. Driven by strong local integrated circuit design capabilities, more than 430 projects focused on lightweight AI architectures to enable real-time processing directly within terminal devices and industrial machinery. Furthermore, 590 teams integrated open-source environments like GitHub and registered on Crunchbase, significantly elevating the international visibility of the technological ecosystem.Extended Business Matchmaking Initiatives Showcase Commercial Achievements at COMPUTEXTo foster substantial commercial partnerships, organizers featured a dedicated industry matchmaking zone during the finals to link venture capital with startup resources. This promotional effort extended into the COMPUTEX exhibition period, where exclusive matchmaking sessions successfully showcased the winning projects to connect the teams with international buyers and global investors. The Ministry of Economic Affairs plans to continue opening institutional pilot lines to support design verification, prototyping, and unified system integrations, ensuring businesses can seamlessly implement artificial intelligence without developing systems from scratch.The Best AI Awards celebrate global excellence in artificial intelligence and IC design, welcoming submissions from innovative companies and brilliant student teams. Following the success of the 2026 edition—advised by the MOEA, organized by DoIT, and executed by TCA—the prestigious competition is officially transitioning into an annual tradition.Offering substantial grand prizes and unmatched industry exposure, the countdown to Best AI Awards 2027 has already begun. Details on the next submission cycle, prize tiers, and eligibility rules will be released soon. Connect with on LinkedIn for the latest official updates and application alerts. Group Photo of the Best AI Awards 2026 Winners. Credit: TCA

Six months into 2026, the story isn't whether the electronics supply chain has stabilized. It has, broadly. The real story is that it has split into two distinct markets moving in opposite directions, and most procurement teams are still planning as if there's just one. The AI-driven leading edge remains capacity-constrained and pricing-positive, while the mature-node segment is loosening into a genuine buyer's market. Knowing which side of that line each line item on your BOM sits on is the single most useful thing you can do heading into the second half of the year.The Two-Speed Market, BrieflyTSMC is reportedly preparing to raise prices 3–10% on its sub-5nm offerings, and with Nvidia and Apple having already locked in large blocks of capacity through year-end, second-tier buyers are increasingly competing for allocation that may not exist in H2. Memory tells a similar story. Combined output from Samsung, SK Hynix, and Micron is expected to grow sharply by 2030, led by a projected rise in HBM production, but that new capacity doesn't meaningfully arrive until 2027. In the meantime, DRAM and HBM remain the tightest categories in the entire component ecosystem.Mature-node wafer pricing has actually returned to pre-pandemic levels, down 5–8% year-over-year, as Chinese fab capacity comes online and automotive/industrial utilization climbs into the 80–85% range. Consumer electronics demand is stabilizing too, which is easing pressure on mature-node semiconductors and passives.What This Actually Means for Your Procurement StrategyThis is where most H2 outlooks stop short. Here's what to actually do with this picture, category by category.1. Segment your BOM by risk profile, not just by part number.Advanced logic tied to AI-adjacent applications faces a fundamentally different supply reality than commodity discrete or mature-node passives. Treat these as two separate procurement strategies, not one blended approach. A BOM review that groups parts by constrained, stable, or loosening rather than by function or supplier will surface where your actual exposure sits, and it's often not where teams assume.2. Use the mature-node buyer's market now, not later.If you were forced into single-sourcing during the 2021–2023 shortage years, H2 2026 is the window to qualify second sources for those mature-node components while pricing and availability both favor you. This window won't stay open indefinitely. As automotive and industrial utilization keeps climbing toward capacity, the leverage shifts back to suppliers.3. Be strategic, not reactive, on memory buys.Memory is not one uniform story. Leading-edge densities tied to HBM demand carry a real price premium, but previous-generation DDR4 and LPDDR4 may still offer value while pricing is structurally supported mainly at the high end. If your designs can tolerate a prior-generation module, this is the year to lock it in rather than wait and hope for relief that isn't coming until 2027.4. Watch for new procurement categories forming in real time.Optical networking components for AI data centres are moving from niche to mainstream as AI clusters push toward much higher bandwidth per rack. If your roadmap touches high-bandwidth AI infrastructure at all, get ahead of this now. Procurement categories that don't exist yet in your sourcing playbook have a way of becoming urgent overnight once a design win locks them in.5. Rebuild safety stock around true risk, not blanket buffers.Broadly increasing inventory across the board is expensive and imprecise. The more effective move is targeting buffers at the small percentage of components, often a single connector, capacitor, or legacy memory module, that actually drive downtime risk if they disappear. Combine that with forecasting discipline: suppliers increasingly prioritize allocation and pricing based on how credible and consistent your rolling forecasts are, which means forecast accuracy is now a negotiating asset, not just a planning exercise.6. Don't sleep on non-obvious demand drivers.Rising defense spending across Asia-Pacific is quietly adding a new, non-cyclical source of demand for industrial and mature-node electronics, one that doesn't show up in most consumer-electronics-driven forecasts. If your end markets touch defense, aerospace, or industrial automation, factor this in as upside demand pressure, not background noise.The Bottom Line for the Rest of 2026H2 2026 doesn't call for a single supply chain strategy. It calls for two, running in parallel. Where you're constrained, the priorities are allocation planning, forecast credibility, and locking in what capacity you can. Where you're not, the priority is using the current leverage to diversify, qualify alternates, and rebuild resilience before the window closes. Teams that treat this as one undifferentiated “stabilizing market” will miss both opportunities.If you're navigating either side of this, securing allocation on constrained parts or taking advantage of loosening mature-node availability, that's exactly the kind of sourcing challenge our team works through with customers every day. Reach out to your Fusion Worldwide representative and let's talk through your BOM.(Article Sponsored by Howard Tan, Director of Purchasing, China Fusion Worldwide)

Pursuing a PhD might have been the most lucrative decision that Tomasz Matusiak has ever taken.Whilst studying at the Wroclaw University of Science and Technology in Poland's third-largest city, he developed chemical sensors made from ceramic materials based on microplasma generators, and electrical components made from a paste of glass and graphite. Now, Matusiak is using this research to solve a bottleneck that plagues the cutting edge of AI development: moving data between where it is stored in memory and where it is handled in the processing unit (for example, the central processing unit [CPU], which handles arithmetic and logical operations; or more specialised graphics processing units [GPUs], which handle computer graphics and digital images) wastes both time and (electrical) power. This limits the extent to which AI models can be scaled up and, of course, harms the environment.What if one could perform all of the computational tasks right where the data are stored? Matusiak thinks his material can do this, and he has started a company, SemiQa, and produced a system inspired by the human brain, the Analog Neural Network (ANN). Unlike conventional chips, which can reach 80°C and require a cooling system, his ANN system only reaches a maximum of 40°C. SemiQa's goal since its inception at the start of 2025 has been to conquer the universe of data centres, replacing their graphic cards (and the GPUs that power these graphic cards) with ANNs.Matusiak wants to bring back analog processing for its computational advantages. He uses the analogy of a train ride through the countryside. One might look outside the window and see a forest pass one by, followed by a short section alongside a river, before heading back into the forest again. A human brain – the analog system – would see a forest and then not think about it again until it sees a change in the environment (the river), and then once again not actively register the river again until the river has been replaced by the forest. It only processes the changes.But a digital system would constantly process what is outside the window. Analog processing thus saves on energy as it doesn’t process when there hasn't been any change.Likewise, digital processing might allocate a large number of bits to a small integer – for example, even though the number 5 can be expressed in binary with just three bits (101), it might be stored in an 8-bit or a 16-bit structure, where most of the surplus bits are zeroes. Many of the operations performed on these small integers will also result in small integers, so most of the leading zeroes will not change. A lot of memory is wasted.Analog processing can get around this problem by simply storing the 5 in a memory cell as a 5 instead of in eight memory cells as 00000101. (Analog memory cells, unlike digital memory cells, can take on more values than just 0 and 1.)The neural approach is based on a special electrical component called a memristor (short for memory resistor). A traditional resistor follows Ohm's Law, which states that the current (the rate at which electric charge flows) through a conductor is proportional to the difference in voltage (or the difference in electric potential energy, or the work it would take to move a unit of charge provided by, for instance, a battery) across that conductor. Mathematically, Ohm's law is V= IR, where V stands for the voltage, I for the current, and R for the resistance of the conductor, a proportionality constant that indicates how difficult it is for charge to move. The higher the resistance, the lower the current (for a given level of voltage).In a traditional resistor, the resistance doesn't vary with current (or voltage). In a memristor, though, the resistance depends not just on the current (or voltage) but also on the past levels of current running through it (or voltage controlling it). In other words, if the voltage goes up and then goes back down to its earlier level, the current and resistance might not return to their original levels. This ability to take on a range of values of resistance also mean that the memristor can be analog – in other words, that it can represent a range of values and not just a 0 or a 1.In addition to its superior thermal properties, SemiQa's ANN1000 is more power-efficient than other chips, being able to carry out more than 30 TOPS (or 30 trillion operations per second) per Watt of power; standard GPUs or NPUs (neural processing units, which are specialised for AI applications) can only carry out 1-2 TOPS per Watt. (The chip consumes 2.5 Watts of power, and so can carry out roughly 75 TOPS.)It is also naturally faster – ANN1000's latency (the time delay between when the processor requests something from memory to when the processor retrieves it) is 50 times shorter than that of a conventional GPU or NPU.The next step is, of course, commercial-scale production of their chips. They already demonstrated a proof-of-concept of their memristive technology at last year's SEMICON Taiwan, an annual trade show and Asia's largest semiconductor event. They will now create a neural network on silicon and hope to have a product-ready chip tailored to specific applications by the end of 2027. The memristive material, a mixture of organic and inorganic parts, is compatible with CMOS (complementary metal-oxide-semiconductor) technology, which is commonly used in foundries to fabricate chips. Matusiak envisions SemiQa's chips in mission-critical applications where efficient power consumption and processing is highly advantageous. These include autonomous systems, such as drones to be used in war and marine robots. Electric cars can also benefit: GPUs currently account for roughly half the cost of driverless vehicles, and replacing conventional GPUs with SemiQa's chips could reduce the price for consumers whilst maintaining manufacturers' margins. SemiQa also plans to add B2B applications such as data centres to the aforementioned B2C applications. They will tackle this through the ANN2000, a matrix of a thousand smaller ANN1000s.The prize money from the Best AI Awards pales in comparison to the 3 million EUR in pre-seed funding that SemiQa has already raised in Europe. But Matusiak is most grateful for the recognition that the judges have given his company's achievements since they started it just a little more than a year ago. This will also facilitate their collaboration with potential partners – in fact, they are already in talks with two local foundries to deepen their co-operation and scale up production of their chips."If you want something special, you need to work with the special forces," says Matusiak. "Everyone knows that Taiwan is the best in the world."SemiQa also plans to set up a branch office in Taiwan and will potentially hire two business developers in the country in the short term. They also know that they will need more funding, and are looking into perhaps raising money from Taiwanese investors. SemiQa already has a strong relationship with Taiwan, being a member of the Taiwan-Poland Chamber of Commerce and having signed memoranda of understanding with several Taiwanese businesses.The Best AI Awards celebrate global excellence in artificial intelligence and IC design, welcoming submissions from innovative companies and brilliant student teams. Following the success of the 2026 edition—advised by the MOEA, organized by DoIT, and executed by TCA—the prestigious competition is officially transitioning into an annual tradition.Offering substantial grand prizes and unmatched industry exposure, the countdown to Best AI Awards 2027 has already begun. Details on the next submission cycle, prize tiers, and eligibility rules will be released soon. Connect with us on LinkedIn for the latest official updates and application alerts. 

By now, it's common knowledge that artificial intelligence has sped up coding and is now replacing junior software engineers. But can you really trust the code that AI produces?The answer is a clear no. Programmers are still needed to review the code for bugs of all sorts. Maybe the code hallucinates, meaning that it has the right syntax and looks plausible but attempts to, for example, import a library that doesn't exist. Maybe there are functional bugs, where the code simply produces the wrong result because of a faulty algorithm and these are just the simple errors.Moreover, companies are naturally hesitant to release sensitive or proprietary information, such as customer data, to an AI coding tool.Inferara wants to solve these problems. It wants to give you code on which you can rely, free from any vulnerabilities, and it does this by reformulating the code into mathematical notation. This is all done on the user's own computing infrastructure, so that no code or data is ever shared externally.Georgii Plotnikov is the CEO and CTO of Inferara. Originally from Russia, Plotnikov had been working as a software developer for a decade after graduating from university. Roughly five years ago, he started learning more about static code analysis, which can be likened to a real-time spell-checker for code: it is, essentially, early detection for any bugs or security vulnerabilities in the code.He started developing an idea for a new way to turn static code analysis into a mathematics problem at the end of 2023. Around the same time, he started to explore where he could settle down and legally incorporate his company. Plotnikov decided on Japan, sent his business plan and other documents to the country,s government, and after being awarded a startup visa, moved there in the summer of 2024. (Inferara is a relatively young company, incorporated only in October 2024.)Unlike many AI companies, which focus on their product and want to get them out to market as quickly as possible, Inferara is proud to call itself a research-first company. Its main product is not the code checker but rather Inference, which he describes as "a programming language that uses mathematics to ensure code works exactly as intended".This is also known as formal verification, and being formally verified gives programmers and users an additional layer of assurance that, for instance, cryptographic protocols will keep data secure, or that compilers for programming languages will not make mistakes. It's also important for mission-critical processes, such as those in automated (driver-less) driving systems, financial systems, energy plant controls, and medical devices.In a nutshell, formal verification examines all possible states that the (usually finite number of) variables or parameters can take, in order to ensure that the logic is sound. In contrast, simply testing the system with common values that the variables might take might not reach all possible states, especially as the number of variables increases, and it will definitely not reach all possible states when there are an infinite number of variables (or values that the variables can take).Mathematical proofs can group variables together and reduce the total number of states that need to be explored. They can even use mathematical induction to verify an infinitely large number of states.Traditionally, programmers have needed a sophisticated understanding of mathematical logic to manually perform formal verification. Inferara's Inference programming language performs formal verification as one codes (or as AI codes) and makes it possible for even those with no understanding of mathematical logic to guarantee their code's fidelity. Inference's diagnostics are also developed so that both humans and AI coding agents can interpret them and, thus, use them to improve the code's reliability.Inference builds upon and improves the user experience of using Rocq, a piece of software that can assist in mathematical proofs. Plotnikov says that Inference sets itself apart by working what appears to be slowly. Instead of writing code as quickly as possible, it continually checks the code's correctness as it is being generated. This takes more time but, in the end, is more cost-effective, since it requires fewer calls to AI later on. Plotnikov likens this to the slower, more rational and conscious system of thinking described in Daniel Kahneman's bestselling book Thinking, Fast and Slow.Inferara has now taken the next step beyond Inference: It has created Vibe Checker, an operating system that can encapsulate the entire AI-led code-development process. In addition to its own desktop AI agent, Vibe Checker is compatible with the existing, widely-used coding agents of Claude Code, Gemini, and Codex.Most companies currently using AI-assisted coding typically do so via the cloud (i.e., off-premises). Although the code and data is stored under the programmer's domain, all that information traverses multiple external barriers when being run. Vibe Checker provides its own gateway, comprising its own server and the user’s own cloud computing security model (also known as bring-your-own-encryption, bring-your-own-key, or BYOK). Vibe Checker's most data-secure services will host everything on users' own infrastructure, and is appropriate for those users who have the resources to buy their own hardware. There is total data sovereignty: no data ever leaves the user's infrastructure, nothing is ever uploaded to Inferara's servers, and everything can be run even without internet access. An audit trail also logs every action taken by a coding agent.Vibe Checker does more than just check code. It's a multifunctional checker, or task-agnostic, and can audit any AI-generated system for functional errors and security vulnerabilities. Law firms can use it to summarise case files and contracts; researchers and policy analysts can use it to verify their database queries; finance departments can check their tax filings; hospital clinicians can manage their patient records; and banks and insurers can double-check their regulatory submissions.Privacy is essential in all of these aforementioned applications, and Vibe Checker enables on-premise analysis with links and citations referring back to the original source of the information – in other words, no hallucinations.Plotnikov says that Inferara will clearly benefit from the 500,000 NTD prize money from the Best AI Awards. However, the most important reward is the valuable feedback from those working in the industry, telling them that what they are doing makes sense and has commercial value. The award will also help Inferara get its name out and allow them to approach and demonstrate their technology to potential partners; in fact, Inferara is currently in discussions with several companies in Taiwan. In particular, they are interested in working with Taiwan's semiconductor and chip industry, which can provide the hardware supply chain for their product.In the next few months, they plan to deploy Vibe Checker to several companies who will serve as beta-testers, including several Taiwanese companies. They also intend to establish a legal entity in Taiwan, and have plans to work with the Centre of Industry Accelerator and Patent Strategy (IAPS) at the National Yang Ming Chiao Tung University.The Best AI Awards celebrate global excellence in artificial intelligence and IC design, welcoming submissions from innovative companies and brilliant student teams. Following the success of the 2026 edition—advised by the MOEA, organized by DoIT, and executed by TCA—the prestigious competition is officially transitioning into an annual tradition.Offering substantial grand prizes and unmatched industry exposure, the countdown to Best AI Awards 2027 has already begun. Details on the next submission cycle, prize tiers, and eligibility rules will be released soon. Connect with us on LinkedIn for the latest official updates and application alerts.