File size: 2,619 Bytes
4f8feb6 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 | # Roadmap
What is done and what is next. The reasoning and status behind it is in [RAG_GAP_ANALYSIS.md](RAG_GAP_ANALYSIS.md). Back to the [README](../README.md).
## πΊοΈ Roadmap
- [x] **BM25 hybrid search** β sparse + dense fusion via Reciprocal Rank Fusion; `vector | bm25 | hybrid` mode per query
- [x] **PDF / DOCX / HTML ingestion** β document loaders with clean text extraction
- [x] **Streaming answers** β server-sent events on `POST /query/stream`
- [x] **Eval regression guard** β `scripts/check_eval.py` in CI; build fails if recall drops
- [x] **PR eval report** β sticky PR comment with metric deltas on the v2 eval set; fails on a drop over 0.01
- [x] **pgvector adapter** with namespaces behind a `Store` interface (Qdrant / Pinecone still open)
- [ ] **Qdrant / Pinecone adapters** β swap the in-memory store for a production vector DB without touching engine callers
- [x] **Cross-encoder reranker** β second-stage BGE reranking via fastembed (`LUMEN_RERANKER`)
- [x] **Layout-aware ingestion** β Docling backend for PDF/DOCX/PPTX/HTML (`LUMEN_DOCLING=1`)
- [x] **Answer-quality eval** β DeepEval faithfulness / relevancy metrics (`lumen eval-answers`)
- [x] **Claim-level diagnostics** β `lumen eval-answers --claims` attributes wrong answers to retrieval vs generation, offline
- [x] **Multi-tenant namespaces** β `LUMEN_NAMESPACE` isolates collections per user/team (pgvector backend)
- [x] **Async FastAPI endpoints** β non-blocking model calls on a pooled async client, blocking work in worker threads (async DB and embedder clients are still open)
- [x] **Real embedder benchmark** β local bge-small (fastembed) next to the hashing baseline; OpenAI row pending
- [x] **Abstention** β confidence-gated refusal, calibrated on held-out questions, measured end to end
- [x] **Answerability check beyond similarity** β an extractive SQuAD 2.0 reader refuses near misses that similarity lets through
- [x] **Query decomposition** for multi-hop questions: measured, no gain with a real embedder (see [benchmarks](benchmarks.md)); kept opt-in
- [ ] **LLM query rewriting** for implicit multi-hop questions (needs an API key with credit)
- [x] **Contextual chunk headers** β offline title + section headers before embedding and BM25; helps keyword search, small effect on dense (see [benchmarks](benchmarks.md))
- [ ] **LLM-written chunk context** β Anthropic's full version, needs an API key
- [ ] **Public benchmark** β BEIR SciFact alongside the in-repo sets
- [ ] **Fine-tuned embedding adapter** β LoRA adapter trained on domain Q&A pairs to improve recall
|