Ritambharam/medvectorai-reranker

Render key points as markdown bullets

Fix collapsed outputs and remove repeated filler key points

Fix instruction echo with natural-language prompt

Simplify generation flow to clean context and single retry

Enforce reasoning-first structured answer generation

Relax summary filtering and enforce answer completeness

Add rule-based precleaning before summary generation

Use fast HF defaults and limit summary workload

Enforce validated summary-first answer generation

Add intermediate chunk summarization before final generation

Add intent-aware context and answer post-filtering

Upgrade generator to Zephyr with FLAN fallback

Improve abstraction and relevance with smart context filtering

Restore full-chunk context and reduce over-filtering

Improve answer quality with context cleaning and robust fallback

Prevent heading-only template outputs

Sanitize template echoes from generated answers

Enforce strict markdown answer sections and source limits

Output markdown sections for premium frontend rendering

Reduce insufficient responses with relevance-aware fallback

Improve full-RAG answer quality and output structure

Align full-RAG prompt, top-8 context, and sources schema

Upgrade HF API: add /embed and /answer with flan-t5 generation

Fix: use valid HF colorTo value

Fix: use CPU-only torch, add HF Space front-matter

Deploy MedVectorAI reranker service

initial commit