feat(rag): switch to rich context judge with reasoning (D3 strategy)

tune(rag): calibrate relevance filter for raw logit scores

feat(api): replace Gemini generation filter with score gap + LLM fallback

feat(rag): add adaptive relevance filter (score gap + LLM fallback)

fix(rag): only rerank semantic queries, skip filter intents

feat: add LRU-cached embed_query to avoid re-embedding repeated searches

feat(rag): integrate cross-encoder reranker into hybrid retrieval pipeline

chore: add missing __init__.py to all packages

chore: remove unused _drop_distant method from retriever

feat(eval): include expected IDs and missed UIDs in eval results

feat(eval): add display module for console formatting and diffs

feat(eval): add runner module for eval orchestration and JSON storage

feat: search catalogue UI, data cleanup, router decade detection

refactor: remove streaming generator and rewriter

feat(api): replace SSE chat with sync POST /api/search

feat(ui): view transitions, story grid 2x2, logo links, scroll UX

feat(mosaic): horizontal scrolling rows, random order, click-through links

fix: pass token_type_ids to ONNX model when expected

feat: correct story links, loading UX, slug-based card filtering

feat: conditional query rewriting for conversational follow-ups

feat: metadata routing for structural queries (most awarded, newest, oldest)

feat: integrate cross-encoder reranker into hybrid retriever

feat: cross-encoder reranker with ONNX inference

feat: suggest related topics when no stories match

feat: show cards only for stories Gemini actually cites

feat: LLM-as-judge relevance filter for story cards

fix: reduce media_type soft boost to prevent multi-facet query issues

fix: remove media_type from hard filter intents

feat: /api/chat SSE endpoint with hybrid RAG

feat: /api/stories endpoint for landing mosaic

feat: FastAPI skeleton with health endpoint

feat: hard filters for structured intents (awards, dates, media_type)

refactor: remove cross-encoder reranker, RRF handles ranking alone

feat: replace sentence-transformers with ONNX Runtime embedder (bge-small 384d)

feat: streaming responses with Flash, remove story cards

feat: retrieval evaluation script with router improvements

fix: add dotenv loading, enrichment retry logic, and graceful error handling

fix: pipeline cleanup + retrieval improvements

fix: address code review issues (theme/css, reranker condense, cli serve)

feat: Gradio UI with hybrid retriever, filter accordion, updated examples

feat: update prompts for stories-only context with enriched metadata

feat: rewrite retriever as HybridRetriever with dense+BM25+RRF+reranking

feat: add QueryRouter with regex-first intent classification

feat: add Reranker with cross-encoder ms-marco-MiniLM-L-12-v2

feat: expand chunk_story metadata (20+ fields), remove chunk_segments

feat: add BM25Store with tokenization, stemming, save/load

refactor: remove segments collection from VectorStore, add get_all_metadata

feat: expand enrichment prompt with themes, countries, event_period, media_type, is_subject + cache invalidation

feat: expand StoryEnrichment with themes, countries, regions, event_period, media_type, is_subject

feat: enrich pipeline story dict with awards, director, content_type, year/decade