How campusx-official/rag-papeer-project works
Papeer is a Streamlit research-paper Q&A app. Users create chat sessions, add files or arXiv papers, then ask questions; the app streams answers from a LangGraph workflow that can retrieve paper passages, assess relevance, rewrite a query, verify claims, or answer directly. The app also offers a /btw side channel for general questions and optional live web search. Uploads are chunked before indexing in session-specific Qdrant collections. Workflow routing details are only partly excerpted, so internal graph transitions are not expanded here.
Main entry points: app.py Streamlit app · rag_graph.py RAG graph · btw_handler.py BTW handler · vector_store.py Vector store
Components 11
Chat and sessions
- Streamlit app
app.py - Chat sessions
app.py
Paper ingestion
- Paper loading
backend/paper_loader.py
Research workflow
- RAG graph
backend/rag_graph.py - Decision models
backend/models.py - BTW handler
backend/btw_handler.py
Index and persistence
- Vector store
backend/vector_store.py
Other components
- Researcher
- Qdrant
- OpenAI models
- Tavily search
Connections 14
- Researcher → Streamlit app: asks or uploads
app.py - Streamlit app → Chat sessions: loads and saves
app.py - Streamlit app → Paper loading: loads papers
app.py - Streamlit app → Vector store: adds and lists
app.py - Streamlit app → RAG graph: streams queries
app.py - Streamlit app → BTW handler: routes /btw
app.py - RAG graph → Vector store: searches passages
backend/rag_graph.py - RAG graph → Decision models: uses decisions
backend/rag_graph.py - BTW handler → Decision models: structures routing
backend/btw_handler.py - BTW handler → OpenAI models: routes and answers
backend/btw_handler.py - BTW handler → Tavily search: searches when needed
backend/btw_handler.py - RAG graph → OpenAI models: routes and answers
backend/rag_graph.py - RAG graph → Tavily search: searches web
backend/rag_graph.py - Vector store → Qdrant: stores vectors
backend/vector_store.py