fix: batch ChromaDB upserts to stay under OpenAI 300k token/request limit

fix: add /debug/backend endpoint and bg error tracking to diagnose upload failures

fix: background upload to avoid HF proxy timeout on large OCR PDFs

fix: add 120s JWT leeway for Clerk dev tokens (60s expiry)

fix: add auth diagnostics, fix CORS origins, show 401 error detail

fix: delete by IDs to avoid ChromaDB compound-filter bug, show trash icon

feat: add JWT auth, per-user document isolation, and delete endpoint

fix: correct HF Spaces URL with domain hash suffix

fix: use root-anchored .vercelignore patterns, revert vercel.json rootDirectory

fix: set rootDirectory=ui in vercel.json to isolate frontend build

fix: add .vercelignore to exclude Python backend from Vercel

feat: add Vercel env config for HF Spaces API endpoint

chore: prepare for HuggingFace Spaces deployment

feat: Update API_BASE configuration and add Docker support

feat: Implement Notebook functionality with tabbed interface and document summarization

Add README.md with full project documentation

Initial commit: Production-Grade RAG with Hybrid Search, Re-ranking and React UI