Hackathon strategy

The main app stays product-first.

This page keeps the internal lane choice out of the demo surface. The product pitch is Scout: the patch tournament for AI-written code. The build strategy is Agentic Coding plus Building Evals.

Pathway choices

Use this for planning and judging narrative, not as the first screen of the app.

Agentic Coding + Evals

Recommended

Specialist scouts evaluate AI-written code, a judge dedupes findings, repair agents compete, and a receipt captures the proof.

tool callsparallel scoutspatch tournamentreceipt

Multimodal Intake

Available

Voice or screen-recorded bug reports can feed the same evaluation pipeline after the core loop is stable.

audiotranscripttriage

Domain Agents

Available

The chosen domain is agentic engineering itself: review infrastructure for code produced by coding agents.

AI-code taxonomypolicyevidence

OpenAI API usage

Seeded demo mode is deterministic for recording reliability. Live repo review uses the OpenAI API in /api/review, and live patch generation uses the OpenAI API in /api/fix. Once hackathon credits are available, set OPENAI_API_KEY and optionally tune OPENAI_MODEL. Scout keeps static prompt rules ahead of dynamic repo context and shows cached-token usage when the live model response reports it.