How priyan-19/research_paper_analysis works
This is a research-paper assistant: users upload academic PDFs, which are parsed, enriched with metadata, chunked, embedded, and indexed for later search and question answering. Its hybrid RAG path combines vector retrieval with graph traversal, fuses evidence, and generates cited answers. Other documented capabilities include graph construction and link prediction, paper analysis, evaluation, and a Streamlit dashboard alongside the REST API. Edges reflect shown calls, imports, or README-level architecture; some endpoint and analysis internals were not sampled.
Main entry points: paper_service.py Paper management · vector_store.py Vector search and index · hybrid_retriever.py Hybrid retrieval and RRF · chat_service.py Grounded chat
Components 25
User interfaces
- Streamlit dashboard
backend/streamlit_app.py - REST API
backend/app/api/v1 - API router
backend/app/api/v1/router.py
Paper processing
- Paper management
backend/app/services/paper_service.py - PDF extraction
backend/app/ingestion/pdf_loader.py - Metadata extraction
backend/app/ingestion/metadata_extractor.py - Semantic chunking
backend/app/ingestion/chunker.py - Paper processing worker
backend/app/workers/paper_processing.py
Search and answers
- Embedding service
backend/app/embeddings/embedding_service.py - Vector search and index
backend/app/embeddings/vector_store.py - Graph retrieval
backend/app/rag/graph_retriever.py - Hybrid retrieval and RRF
backend/app/rag/hybrid_retriever.py - Grounded chat
backend/app/rag/chat_service.py - Vector RAG
backend/app/rag/vector_rag_service.py - LLM providers
backend/app/llm
Knowledge and analysis
- Graph extraction and building
backend/app/graph - Entity canonicalization
backend/app/graph/canonicalizer.py - Link prediction
backend/app/graph/gnn_link_predictor.py - Research analysis
backend/app/analysis - RAG evaluation
backend/app/evaluation/metrics.py
Runtime and storage
- PostgreSQL and pgvector
backend/app/database/postgres.py - Neo4j graph store
backend/app/database/neo4j.py - Redis
backend/app/database/redis.py - Application runtime
backend/app/main.py
Other components
- Researcher
Connections 41
- Researcher → Streamlit dashboard: uses
README.md - Researcher → REST API: requests
README.md - Streamlit dashboard → Paper management: manages papers
backend/streamlit_app.py - Streamlit dashboard → Grounded chat: asks questions
backend/streamlit_app.py - Streamlit dashboard → Hybrid retrieval and RRF: searches evidence
backend/streamlit_app.py - Streamlit dashboard → Link prediction: runs predictions
backend/streamlit_app.py - Application runtime → API router: mounts routes
backend/app/main.py - API router → REST API: registers endpoints
backend/app/api/v1/router.py - REST API → Paper management: manages papers
- REST API → Hybrid retrieval and RRF: searches
- REST API → Grounded chat: answers questions
- Paper management → PDF extraction: loads PDF
backend/app/services/paper_service.py - Paper management → Metadata extraction: extracts metadata
backend/app/services/paper_service.py - Paper management → Semantic chunking: creates chunks
backend/app/services/paper_service.py - Paper management → PostgreSQL and pgvector: persists papers
- Paper management → Vector search and index: embeds chunks
backend/app/services/paper_service.py - Paper processing worker → PDF extraction: loads PDF
backend/app/workers/paper_processing.py - Paper processing worker → Semantic chunking: chunks text
backend/app/workers/paper_processing.py - Paper processing worker → Vector search and index: stores embeddings
backend/app/workers/paper_processing.py - Paper processing worker → Graph extraction and building: extracts graph
backend/app/workers/paper_processing.py - Embedding service → Vector search and index: creates vectors
backend/app/embeddings/vector_store.py - Vector search and index → PostgreSQL and pgvector: searches and stores vectors
backend/app/embeddings/vector_store.py - Hybrid retrieval and RRF → Vector search and index: retrieves vectors
backend/app/rag/hybrid_retriever.py - Hybrid retrieval and RRF → Graph retrieval: retrieves graph context
backend/app/rag/hybrid_retriever.py - Hybrid retrieval and RRF → Graph retrieval: fuses ranked results
backend/app/rag/hybrid_retriever.py - Graph retrieval → Neo4j graph store: traverses graph
backend/app/rag/graph_retriever.py - Graph retrieval → Entity canonicalization: canonicalizes entities
backend/app/rag/graph_retriever.py - Grounded chat → Hybrid retrieval and RRF: retrieves evidence
backend/app/rag/chat_service.py - Grounded chat → Vector search and index: uses vector results
backend/app/rag/chat_service.py - Grounded chat → LLM providers: generates answer
backend/app/rag/chat_service.py - Vector RAG → Vector search and index: retrieves passages
backend/app/rag/vector_rag_service.py - Vector RAG → LLM providers: generates answer
backend/app/rag/vector_rag_service.py - Graph extraction and building → Neo4j graph store: builds graph
backend/app/graph/neo4j_builder.py - Link prediction → Entity canonicalization: uses canonical entities
backend/app/graph/gnn_link_predictor.py - Research analysis → Paper management: uses paper records
backend/app/analysis/similarity_detector.py - Research analysis → Vector search and index: checks similarity
backend/app/analysis/similarity_detector.py - Application runtime → PostgreSQL and pgvector: initializes
backend/app/main.py - Application runtime → Neo4j graph store: connects
backend/app/main.py - Application runtime → Redis: connects
backend/app/main.py - RAG evaluation → Hybrid retrieval and RRF: evaluates retrieval
- Grounded chat → RAG evaluation: measured by