fix: normalize cross-encoder scores to 0-100% range

fix: change Uvicorn port to 7860 for HuggingFace health check

fix: resolve CORS, cache models in RAM, optimize chunk size

chore: add huggingface spaces configuration to readme

fix: load embedding and reranker models from local /app/models path, never HuggingFace Hub at runtime

fix: update Dockerfile to use download_models.py with deterministic /app/models cache path

feat: add download_models.py to pre-cache HuggingFace models at build time

feat: add structured startup logging to improve production observability

chore: pin all backend dependency versions for reproducible Docker builds

feat: add multi-stage Dockerfile for FastAPI backend with model pre-bake

feat: enrich /health response and add /ready liveness probe for Cloud Run

feat: reconcile existing FAISS indexes into session store on startup

feat: add preload_models.py to cache HuggingFace models at image build time

chore: add backend/.dockerignore to exclude venv, indexes, and secrets

chore: externalize RERANKER_MODEL, EMBEDDING_MODEL, GEMINI_MODEL to env vars

fix: secure CORS with ALLOWED_ORIGINS env var, remove unused imports

fix: add missing langchain-core dependency

fix: resolve protobuf dependency conflict in requirements

feat: backend setup with FastAPI and RAG logic