Pinned Loading

1. axiom axiom Public

   Prove every "done." — verifies AI agent completion claims against evidence declared before the work. Observe-mode by default, zero-config, four runtimes.

   Python 4 1

2. ryanai-evalbank ryanai-evalbank Public

   evalbank: a method and harness for scoring a model on your own work, without publishing the questions

   Python

3. ryanai-lab-cookbooks ryanai-lab-cookbooks Public

   Index of the RyanAI Lab cookbooks: five reader paths, nine sections, 54 cookbooks with a current/rollback/historical status

4. training-reranker training-reranker Public

   Self-hosted reranker and the missing-scoring-head GGUF trap that fakes a weak model

   Python

5. agentic-rag-from-scratch agentic-rag-from-scratch Public

   A six-stage agentic RAG loop in ~200 lines of pure-stdlib Python — gap critic, adversarial verification, cited synthesis. Retrieval is a loop, not a lookup.

   Python

6. training-embedding-finetune training-embedding-finetune Public

   Fine-tune your own embedding model: pipeline scripts, Ollama deployment, and three production incidents

   Shell