This directory is an engine parts catalog — each layer represents a deliberate component boundary in the architecture of production-grade AI systems.
Layers are components that compose into an AI agent. The agent itself lives at repo root in ai_agents/ — the assembled product, not another layer.
The Five Layers¶
| Layer | Directory | Focus |
|---|---|---|
| 1 | 1_execution/ | CPU/GPU optimization, VRAM/RAM management, CUDA pipelines |
| 2 | 2_model/ | SLM selection, tokenization, embeddings |
| 3 | 3_prompts/ | Prompts-as-Infrastructure, consultant prompts, token economics |
| 4 | 4_orchestration/ | RAG, agent workflows, structured output |
| 5 | 5_context/ | Vector stores, hybrid retrieval, indexing strategies |
Read more in A Multi-Layered AI System Architecture.