bhanu-marisa/simple-rag

feat(eval): add RAGAS evaluation script with golden dataset

perf(deploy): swap to MiniLM embeddings, remove flashrank to reduce cold start time

chore(ci): update actions/checkout to v4

fix(ui): cache agent loading with st.cache_resource to prevent reloading on every rerun

fix(ui): resolve merge conflict

feat(yaml): add HuggingFace workflow yaml

fix(ui): call agent directly, remove FastAPI dependency for HuggingFace deployment

fix(deploy): Rename README for HF Spaces

feat(deploy): add HuggingFace Spaces support, simplify ui.py to call agent directly

feat(deploy): add render.yaml for Render deployment

feat(deploy): add Railway config, make API_URL configurable for deployment

feat(ui): add FastAPI + Streamlit chat interface

docs(readme): mark level 5 complete

fix(agent): pass history to rewriter, add upfront rewrite, restore metadata filter, remove streaming delay

feat(agent): add agentic RAG with LangGraph - retrieve, grade, rewrite loop

feat(reranker): add FlashrankRerank to re-score retrieved chunks for better relevance

feat(rewriter): add query rewriting for follow-up questions using conversation history

docs(readme): mark level 4 complete, update current level

feat(streaming): stream LLM response word by word

feat(citations): delegate source citations to LLM for accurate attribution

feat(memory): add conversation history across questions in session

docs(readme): mark level 3 complete, update current level

feat(retriever): add metadata filtering with [docname] prefix syntax

fix(docs): clean wikipedia scraper - remove LaTeX, citations and edit tags

feat(retriever): add hybrid search combining BM25 keyword and MMR vector retrieval

feat(retriever): replace similarity search with MMR for diverse chunk retrieval

docs(readme): add full project readme with setup, architecture and roadmap

chore: add wikipedia html-to-text scraper

feat(rag): upgrade to level 2 - real docs, chunking, persistent chromadb

chore(deps): add requirements.txt

feature: better logs, mechanisms for matching

first commit: simple rag app