Honest eval narrative + query rewrite off by default

Honest eval narrative + query rewrite off by default

Honest eval narrative + query rewrite off by default

Honest eval narrative + query rewrite off by default

Add query rewriting + corrective RAG + 3-stage RAGAS ablation

Add query rewriting + corrective RAG + 3-stage RAGAS ablation

Add query rewriting + corrective RAG + 3-stage RAGAS ablation

Add query rewriting + corrective RAG + 3-stage RAGAS ablation

Add query rewriting + corrective RAG + 3-stage RAGAS ablation

Add query rewriting + corrective RAG + 3-stage RAGAS ablation

Add query rewriting + corrective RAG + 3-stage RAGAS ablation

Add RAGAS evaluation results (reranking improves all 4 metrics)

Add RAGAS evaluation results (reranking improves all 4 metrics)

Add RAGAS evaluation results (reranking improves all 4 metrics)

Add evaluation notebook

Add reranking + RAGAS evaluation

Add reranking + RAGAS evaluation

Add reranking + RAGAS evaluation

Add reranking + RAGAS evaluation

Add reranking + RAGAS evaluation

Add reranking + RAGAS evaluation

Fix LLM tone: forbid based-on-provided-texts phrasing

Add pre-built vectorstore: instant startup, no ingest needed

Add test_history.py: multi-turn conversation memory test

Suppress ChromaDB warning for off-topic (low-similarity) queries

Remove Gemini 3 Flash: persistent 503 unavailable

Fix Google genai contents format: parts must be [{text: ...}] not [str]

Fix history error: normalise Gradio list-of-blocks content to str

Cap LLM history at MAX_HISTORY_TURNS=4 to avoid context overflow

Cap LLM history at MAX_HISTORY_TURNS=4 to avoid context overflow

Always enrich retrieval query with previous user message

Add conversation memory: history-aware LLM + context-aware retrieval

Add conversation memory: history-aware LLM + context-aware retrieval

Fix startup: run ingest in background thread so Gradio launches immediately

Fix HF auth: explicit login from HF_TOKEN before gated model download

Deploy: auto-ingest, hybrid RAG, streaming, UMAP viz, 16 LLMs

initial commit