Building a Chess Coach
summary
Production chess coach combines classical engines and detectors with an LLM that only translates structured signals into English: Stockfish handles position evaluation, tactical and positional detectors fire forks pins and structural themes, and the LLM avoids reasoning. Latency target is met with Gemini 3 Flash for sub-3-second latency in the app. User flags post to Slack and get injected into a running Claude Code Channels session where an autonomous agent triages modifies prompts or detectors and regenerates commentary — the closed loop runs and produces deployable PRs from mobile.
AI-written summary · about
chapters (11)
- 0:00Chess Coach project intro
- 1:19Play Magnus and AI features
- 3:03Chess and AI history
- 6:15LLMs struggle with chess
- 7:00Game review pipeline
- 9:49Autonomous agents loop
- 10:11Demo: autonomous agent
- 12:41Latency vs quality trade-offs
- 13:43Model performance evaluation
- 15:48Key learnings takeaways
- 16:51Chess simul event
related talks
01h 14mBuilding Next Gen Agents with MCP
08mMinimax search and the structure of cognition!
48mMastering the AI Shift: Adaptive Strategies for Software Evolution
13mLet’s run a tiny chess neural network by hand!
15mChess in the Age of Artificial Intelligence
26mtalks about being a good AI Coach
30mTeaching Machines To Debug Machines: Building an AI-Powered RCA...
17mPlay Magnus