AITEX Summit Spring 2026 Hackathon Tested AI Projects Through Product Review

For a technical audience, the useful part of a hackathon story is rarely the trophy order. The stronger question is what work was available for review, how projects were compared, and whether the criteria reached past a polished demo. AITEX Summit Spring 2026, which ran from May 30 to June 1 under the theme “From Concept to Creation,” published a structure that makes those questions easier to examine. More than 100 requests to participate were narrowed to six finalist teams competing across Open Innovation, AI for Good, AI for Business, and Intelligent Systems, with projects scored on originality, technical implementation, real-world impact, presentation, and completeness.

A panel of more than 40 specialists from different countries reviewed the work, drawn from technology companies, research institutions, and startups, with credentials checked before judges were seated. That gives the competition several technical lenses for assessing early AI products and keeps the review less dependent on one person’s taste. With only six teams at the finish line, the panel could stay close to product behavior rather than pitch language, asking how each system holds up after the first impressive screen.

Increasingly, that kind of review is about more than reading code. The strongest hackathon panels now weigh whether a team can take direction as much as what it shipped, and several judges were recognized for that work. Pani Siddharth was honored for the hands-on mentorship he gave teams stuck on technical and strategic problems, and Roman Nekrasov for pushing teams past immediate execution toward questions of fit and positioning. Radjabov Rakhmon was recognized for turning feedback into genuine learning rather than a verdict, the difference between telling a team its score and leaving it able to build the next thing better.

TalentMaster: Privacy and Evidence Structure

First place went to TalentMaster, a local-first, private-by-default SwiftUI app for high-skilled professionals, researchers, founders, and creators. It helps users preserve a professional footprint, map achievements to international immigration criteria, and receive AI-powered recommendations for strengthening a portfolio. Because the materials it handles can include sensitive personal records, unpublished work, and immigration files, private-by-default design addresses a risk that marketing language cannot solve after the fact, and the evidence-mapping workflow gave judges something concrete to inspect: whether a recommendation traces back to a record the user controls rather than to a loose text generator.

The architecture is what made the project reviewable, which is where Daria Ishkova came in. Recognized for technical architecture excellence, she pressed on where data lives and what the AI layer actually sees, while Alexandr Lomovtsev, recognized for setting the evaluation benchmark other judges calibrated against, helped hold a first-place result to a bar the rest of the field could be measured by. A record’s product also lives or dies on whether a non-specialist understands it, which is the harder craft, a point reflected in the knowledge-transfer recognition given to Sofia Kalinina, honored for translating production-grade experience into advice that teams could actually use.

MedSafety — Safety Boundaries and Plain-English Explanation

MedSafety, awarded second place, is a personal medication-safety tool for patients managing multiple prescriptions across different doctors. It maintains one current medication list, checks every drug pair for interactions, ranks them by severity, and uses an LLM to explain findings in plain English. The important engineering choice is the order of operations: the list, the pairwise check, and the severity ranking form a safety chain, and the language model sits after those steps to help a patient understand the result rather than generate it. In a health-related project that separation is not cosmetic, and it gave the panel clear failure modes to test, from an outdated list to an explanation that runs past the underlying check and creates false confidence.

Those are reliability questions as much as safety ones, which is why Daniil Romashov, recognized for system reliability and resilience, mattered to this review. He pressed on how the product behaves once the presenter is no longer driving, the latency, brittle integrations, and missing fallbacks that sink early AI products in ordinary ways. Oleg Marushchak, recognized for quality assurance, kept the review consistent so a safety-adjacent tool was held to the same standard as everything else, regardless of how confident its demo looked.

MOLE INFILTRATOR — Agent Behavior Under Constraints

MOLE INFILTRATOR, the third-place project, is a word game where the player faces AI agents, one secretly sabotaging the group. Each round, all agents see a secret word and submit a one-word clue; duplicate clues cancel out, and after four rounds, the player tries to identify the mole. The one-word limit is the real constraint, since long explanations cannot rescue a weak agent decision, so clue diversity, hidden-role behavior, and a workable difficulty curve all have to come through in a single word, and the duplicate-cancellation rule exposes stale or predictable agents fast. That makes the game a compact test of agent design rather than a playful wrapper around a language model.

Judging it well took an eye for behavior under those limits. Pratikkumar Chaudhari, recognized for analytics innovation, was well-suited to a project whose quality lives in patterns across many rounds of agent output, and Sergey Ryabov, recognized for outstanding evaluation, gave the kind of line-by-line review that catches where a clever demo and its underlying build diverge. A game built on hidden roles is also exactly the kind of submission that looks slight at first glance, which is where a harder judging instinct shows: Akbar Sayakov was recognized for spotting product potential in early-stage work that is easy to underrate, the instinct to ask what a rough prototype becomes rather than only what it is on submission day.

Reading Where Projects Go Next

That forward read, trajectory over snapshot, is becoming a part of hackathon judging that separates a useful panel from a ceremonial one. Early AI products are almost by definition unfinished, so the more valuable question is often whether a team has built something with a credible path forward. Ievgen Gartman, recognized for visionary impact, worked with teams on exactly that, helping them see which parts of a prototype were worth carrying into a real product and which were scaffolding to discard, and pressing on whether the technical approach would survive contact with more data and more users. Mykhailo Krasovskyi, a lead solution architect with deep experience in enterprise cloud migrations and microservices transformations, was recognized for predictive leadership, weighing each project not only on its current state but on its long-term technical scalability and market potential, the read that tells a promising direction from a dead end. Andrei Mishurin was recognized for the groundwork his feedback laid for teams to keep building after the event, the kind of guidance that turns a weekend demo into a roadmap a team can act on once the deadline is gone.

Holding that standard across a mixed-track field is its own discipline. Andrii Stetsenko, recognized for cross-track collaboration, gave teams the same depth of review whether their project was a social-impact tool, a health workflow, or an open-ended experiment, so a strong submission was not penalized for sitting in a less crowded track. Oleksandr Orlov was recognized for judging leadership that kept the panel aligned on what each score actually meant, reconciling differences in deliberation so teams received consistent signals rather than contradictory ones. Oleksandr Rotar was recognized for the communication that kept feedback clear and usable in every exchange, the connective work that lets a large panel reach a result teams can trust and learn from.

Aleksandr Burmistrov rounded out the panel with senior engineering depth: more than 20 years in software engineering and over a decade in executive technical management, including building and scaling EdTech platforms serving tens of thousands of students, along with hands-on AI and secure-development work. In a competition with private records, user safety, and applied AI workflows, that mix of scale and security experience was directly relevant to the review.

What the Structure Adds Up To

The event adds up to more than 100 participation requests, six finishing teams, more than 40 international judges, four tracks, five scoring criteria, and three ranked winners. For a technical audience, the most useful conclusion is that the hackathon placed early AI products under a product-review lens. The winners were not merely named; their systems each exposed a different technical question, from privacy and evidence mapping to safety boundaries and patient-facing explanation to agent behavior under tight constraints, that the competition’s rubric was built to examine.