sangwon38383/geoyoung-docs-api

feat: admin dashboard (files + search logs)

feat: Gemini embedding API (drop GPU)

Upload src/retrieval.py with huggingface_hub

Upload src/embeddings_gemini.py with huggingface_hub

Upload config.py with huggingface_hub

feat: OpenAI GPT fallback when Claude overloaded

Upload src/rag_openai.py with huggingface_hub

Upload config.py with huggingface_hub

feat: include document_id in search sources (clickable citations)

feat: search scopes to current folder subtree

feat: original-file storage on R2 + download

feat: email notification on folder share (Resend)

fix: tolerate transient Anthropic errors in search stream (retry + clean message)

feat: folder-level sharing (개인/공유 문서함, email shares, RW)

Processing progress: staged percent+stage callbacks (OCR/embed sliced)

Retrieval: expand each hit with +/-2 neighbor chunks (skip missing at doc edges)

GPU OOM fix: cap max_seq_length=1024, embed batch 32

GPU OOM fix: expandable_segments env

Resolve CUDA device explicitly; expose gpu in /health

Drop LLM summary (keep filename prefix); embed batch 16->64

Prepend filename+summary to chunks; markdown/LaTeX answer prompt

OCR page-split + page numbers + chunked upload

Async background upload processing

Fix chunk id length (Chroma quota) + file delete endpoint + convert timeout

Deploy v2 backend API

Upload Dockerfile with huggingface_hub

Upload README.md with huggingface_hub

initial commit