HamsterGang
Target Audience: This game is designed specifically for teenagers, students, and young adults interested in technology, as well as fans of psychological mystery games and digital trust. It also serves local audiences through Azerbaijani-language support. The Problem: As AI becomes capable of perfectly mimicking human behavior and generating convincing responses, people increasingly evaluate information based on how natural or confident it sounds rather than on evidence. Today, users face severe challenges with AI impersonation, unverified digital trust, and widespread misinformation. They lack the critical thinking tools needed to identify contradictions, evaluate sources, and recognize when natural-sounding interactions are actually machine-generated.
Preliminary score
- Value for the user14.0 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI17.5 / 30
A working core scenario, and what AI actually contributes to the result.
- Quality testing8.5 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility4.5 / 15
Data requirements, running costs and a clear next step.
- Originality6.0 / 10
A meaningful difference from common solutions.
What the judges wrote
- GPT-6 AstraPreliminary53/100
- Strengths
- Your four-suspect interrogation loop gives AI dialogue a meaningful gameplay role: players compare statements before making a consequential accusation. Your Unity scenes, documented controls, and Gemini conversation flow provide concrete implementation detail, while Azerbaijani storytelling adds local relevance. Your reported knowledge leaks, premature spoilers, and latency issues identify important failure modes, and your README responsibly distinguishes the fictional premise from real-world AI detection.
- To improve
- Your submission would be stronger with a captured end-to-end playthrough showing live questions, generated answers, evidence comparison, and the accusation outcome. Your testing should include reproducible prompts and outputs before and after fixes, violation counts, latency measurements, and comparison with scripted dialogue. Your feasibility plan should explain how case facts and knowledge tiers are authored, estimate API cost per session, and identify the next implementation milestone. Your claimed critical-thinking benefit needs player evidence, such as whether players improve at evidence-based decisions rather than merely guessing the hidden AI.
- Claude Opus 5.5Preliminary48/100
- Strengths
- Your team shipped a playable-looking Unity 6 3D game, with an itch.io build, repository and video. The deck screenshots show real interrogation scenes and four modelled suspects. Gemini is central to the core loop: players type their own questions and NPCs answer from their own personality and limited knowledge, so the AI shapes the gameplay rather than decorating it. Your testing write-up names real LLM failure modes, including out-of-character answers, premature clue leaks and latency, and responds to them sensibly with tiered knowledge unlocking and prefetching.
- To improve
- Show real evidence of the AI working. Your deck's conversation is labelled 'illustrative'. Include actual transcripts from the build, ideally one in Azerbaijani, and at least one captured failure before and after your fixes. Turn your testing into something measurable. For example, run 20 adversarial questions per suspect and count how often each one breaks character or leaks a clue, then compare that with a scripted dialogue tree. Address feasibility directly: estimate Gemini cost per playthrough, explain how you will move the API key server-side (Supabase is listed but not explained), and name a concrete next step such as a school pilot that measures critical-thinking outcomes.
The team
Mirkamil Guliyev
designer
Ayaz Babashov
developer
Kenan AhmadovLead
developer