Commit History
-
chore: remove unused data_dir / DATA_DIR (Storage Bucket dependency)
MATSUMURA Taishi +1: Claude Sonnet 4.6Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>841d2e6
-
fix(neo4j): move data dirs from /data (S3 FUSE) to /tmp (local)
MATSUMURA Taishi +1: Claude Sonnet 4.6Storage Bucket at /data uses FUSE/S3, which is incompatible with Neo4j's file locking and random I/O requirements. This saturated container network resources, preventing uvicorn from serving health checks. Moving Neo4j dirs to /tmp (local ephemeral storage) fixes this. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>a8e89f8 -
debug(spaces): HEALTHCHECK + neo4j logs to stdout for HF health check diagnosis
MATSUMURA Taishi +1: Claude Sonnet 4.6- Add Dockerfile HEALTHCHECK (curl /health, start-period=90s) so Docker marks the container healthy after neo4j stabilizes - Redirect neo4j logs from /tmp to stdout/stderr so they appear in HF Container logs - Add flush=True prints in startup and /health to confirm requests arrive Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>c8e5f4b
-
fix(api): ensure health checks always return 200 for HF Spaces
MATSUMURA Taishi +1: Claude Sonnet 4.6Add /health endpoint and move catch-all route outside the STATIC_DIR guard so the app responds with 200 even when frontend/dist is missing, preventing the 30-min launch timeout on HuggingFace Spaces. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>1db9cff
-
refactor(extractor): migrate to Responses API (client.responses.create)
- Use instructions + input instead of messages - json_object format for structured output - max_output_tokens replaces max_completion_tokens Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>067e38e -
fix(extractor): max_tokens -> max_completion_tokens for gpt-5-mini, bump openai>=1.84.0
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>0332e82 -
refactor(extractor): remove heuristic fallback, require OpenAI API key
- Heuristic path was silently hiding missing key errors - extract_triples now raises HTTP 400 if OPENAI_API_KEY unset - Switch default model to gpt-5-mini Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>2b5418e -
feat(sample): switch Load Sample to Japanese supply chain text
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>a2aee7d -
feat(nlp): add Japanese language support
- chunker: split on 。！？ in addition to . - triple_extractor: add Japanese causal patterns for heuristic fallback - triple_extractor: update LLM prompt to preserve source language in cause/effect Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>a8dddda
-
fix(api): run Neo4j init as background task so FastAPI responds immediately
0e412ae -
feat: implement CC-RAG prototype with Neo4j, FastAPI, React
Full single-container app for Hugging Face Docker Spaces: - Neo4j causal graph storage with forward/backward chain traversal - FastAPI backend: upload, triple extraction, graph build, CC-RAG chat - React+Vite frontend: upload, triples table, Cytoscape graph view, chat - Heuristic fallback extraction when no OpenAI API key - LFS rules for images, fonts, PDFs, native libs, DB filese675bdb