multi-model Groq fallback — llama-70b → llama-4-scout → gpt-oss-120b → llama-8b

on demand pubmed lookup

judge agent in UI, evidence quality cards, study breakdown

fix: remove hardcoded paths for HuggingFace deployment

feat: Phase 2 BioBERT embeddings, accuracy 80% → 90%, response time -38%

updated benchmarking stratergy and evaluated results

added clearing the cache script

added phase 1 progress accuracy and all