docs: per-package reference + showcase site (GitHub Pages) (#107)

Gap analysis: reranker jump + 16-question residue dissection (#87)

Reranker as the precision stage + fold glossing into Build Index (#85)

Answer-quality eval (LLM-judge) + hy3-led OpenRouter default (#81)

Update the gap analysis with today's measurements

Docs: retrieval gaps + improvement roadmap (OKF gap-analysis)

Revert the embedder to bge-small: the bge-base A/B showed no gain

Swap the embedder: bge-small (384d) -> bge-base (768d)

M5: Gradio web app + HF Spaces deploy + live-hydrate fallback

M3.4: LangGraph rewrite of the agent + loop-vs-langgraph comparison

Switch embeddings to a local CPU model (bge-small): free tier APIs can't index the corpus

Choose Gemini free-tier embeddings (gemini-embedding-001)

State the dual mission and declare per-user-version answers a non-goal

Adopt always-latest policy: track /docs/stable, drop version pinning

Automate embedding refresh: release watcher + hash-diff, not commit tracking

Adopt agent-with-tools architecture; integrate DeepWiki for source questions

Expand corpus to the full docs.pytorch.org umbrella: tiered seed list

Drop the code-execution sandbox; define the product as a docs Q&A assistant

Rescope corpus from pinned source modules to the pytorch.org docs site

Add design doc: content pipeline, agent access, session flow, orchestration, live links

Scaffold repo structure per M0 (pyproject, ruff, pytest, pre-commit)