A production repo blueprint for LLM apps: nine directories, clear responsibilities
techNmak · x · 2026-10-08
A detailed repository blueprint for production AI applications, organizing engineering responsibilities around an LLM into api/, application/, providers/, security/, prompts/, observability/, evals/, tests/, and deploy/ directories, with RAG, agents, caching, tool execution, persistence, and workers as optional capability layers. Key argument: prompt changes can degrade quality without breaking unit tests, so structure must make tracing, evals, and permission enforcement first-class concerns. Start small and add capabilities as the product needs them.
Related event: A Production-Grade Repository Blueprint for LLM Applications(2 posts)→
More from coding & agent
- AI agents made writing code cheap, but reviewing generated code is a different job — alex_verem · 2026-10-08
- CADFather: Autonomous Agentic System Reconstructs Parametric CAD from 3D Meshes — Gennadiy Savrasov · 2026-10-08
- UniSkill: Actor-Aligned Skill Proposals Hit 98.4% Success on ALFWorld — Yifei Lu · 2026-10-08
- Client banned AI from payments data; 15 months later they asked for the AI setup — alex_verem · 2026-10-08
- Open-Source FrameForge Chains Prompts on MiniMax h3 for Long Combat Videos — Super_Range45 · 2026-10-08
- Aseprite MCP Rebuild: Grid Loop + Pixel-Art Knowledge Base Boosts Output Quality — Complex-Log-635 · 2026-10-08