Format load test script

Expose cache header observability

Improve validation and resource handling

Add Qdrant retry logic

Replace manual dict construction

Simplified cold-start, API takes query directly

Use canonical is_refusal() from faithfulness module

Replace custom percentile with np.percentile

Extract magic numbers to constants incl other polishing

Harden summary.py

Calibrate evidence quality gate thresholds

Add mixed evidence handling to explanation prompts

Add bootstrap confidence intervals to evaluation metrics

Refactor sanity checks with improved adversarial detection

Track rated_count incrementally

Add concurrent load testing for SemanticCache

Add RefusalType enum and query sanitization

Reorder index creation before upload

Eliminate duplicate payload.get() calls

Fix EDA review-level rating distribution and resource cleanup

Consolidate API reference endpoints

Fix README accuracy and track Kaggle notebook

Improve README accuracy

Fix eval workflow and align README metrics with eval_results JSON

Add .hfignore to exclude binaries from HuggingFace deployment

Fix EDA figure paths for GitHub display

Fix hardcoded HHEM metric

Restructure Makefile

Replace EDA with production Qdrant queries

Switch deployment from Render to Hugging Face Spaces

Switch deployment from Render to Hugging Face Spaces

Switch deployment from Render to Hugging Face Spaces

Add docker-compose, multi-stage builds, and developer tooling

Add docker-compose, multi-stage builds, and developer tooling

Fix RAGAS faithfulness scoring (0.64 → 0.82)

Apply ruff formatting

Fix all CI lint and type errors

Initial commit: Sage RAG recommendation system