Synapse Game
Target User: Indie developers and QA leads at AA/AAA studios. We built this SDK specifically for production teams shipping under tight deadlines—where a post-launch bug directly tanks retention and revenue. It isn't a general-purpose tool, and it isn't for everyone. The Problem: Catching collision and clipping ("wall-walking") bugs manually takes hundreds of hours per build and drains studio budgets. Human testers inevitably miss things, especially around invisible mesh boundaries. When those slip into production, you get brutal Day-1 launches, review bombing, and costly emergency hotfixes (think Cyberpunk 2077). Existing automation macros don't solve this either—they run rigid, linear pathing that completely fails to mimic the chaotic, multi-axis movements of real players, leaving games wide open to exploits.
Preliminary score
- Value for the user16.5 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI13.0 / 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.
- Feasibility7.0 / 15
Data requirements, running costs and a clear next step.
- Originality6.0 / 10
A meaningful difference from common solutions.
What the judges wrote
- Claude Opus 5.5Preliminary49/100
- Strengths
- Your target user (QA teams hunting collision and clipping bugs under release pressure) is concrete, and the pain is real and well articulated. The Unity console screenshot shows an actual in-engine capture loop logging a collision event and saving telemetry. Your honest account of the false-positive problem (foliage, alpha textures, curtains) and the physics layer-mask filter you added before calling the Vision API is a genuine, specific failure-and-fix story. Keeping assets local and sending only telemetry and frames is a sensible design choice for studios worried about IP.
- To improve
- Your materials show little of what GPT-4o-mini actually contributes. Detection is a deterministic heuristic (velocity near zero for 2s), so show a real generated Jira ticket and explain why the LLM output beats a templated report. Replace claims like '100% reliability' and 'near-zero false positives' with numbers: how many broken colliders you planted, how many were caught, missed, or falsely flagged, before and after the layer-mask fix. Compare that against the time a manual tester needs for the same scene. Clarify how a stuck-bot heuristic detects pass-through clipping bugs, and fix the leftover template text on the AI slide about 'pests and diseases'.
- GPT-6 AstraPreliminary53/100
- Strengths
- Your focus on collision QA addresses a concrete problem for game developers, with incident telemetry and bug reports as useful outputs. Your Unity console screenshot supports the detection/capture stage, although it does not establish the full AI reporting pipeline. Your disclosure of foliage and curtain false positives identifies a specific failure mode and a relevant physics-layer filtering response.
- To improve
- Show one end-to-end run from NPC exploration to the generated ticket, explaining the exploration policy and what GPT-4o-mini adds beyond deterministic C# checks. Your tests should distinguish normal wall blocking from true clipping and report case counts, false positives and missed bugs before and after filtering, plus a scripted or manual QA baseline. Reconcile your deck's local-only claim with screenshots sent to OpenAI, and substantiate the $0.05 per 1,000 logs estimate with API-call frequency and image/token costs. Make your next step a small external Unity pilot measuring bug-finding time and coverage.
The team
İbrahim Qurbanov
designer
Tural PiriyevLead
developer
Qasid Nerimanli
developer