How aryansaraogi/researchrag works
ResearchRAG lets researchers import papers, search and ask citation-grounded questions, compare papers, explore citation links, and evaluate answer and retrieval quality. Its main workflow is paper ingestion into searchable chunks, followed by query planning, filtered hybrid retrieval, answer generation, and citation mapping to paper pages. The README documents several capabilities whose implementation details were not sampled; those branches are shown at the subsystem level. Edges use excerpts or README statements as evidence; unsampled internals are not inferred.
Main entry points: pipeline.py Ingestion pipeline · layout.tsx Research UI · hybrid.py Hybrid retrieval · rag.py Query planning and RAG
Components 34
Paper ingestion
- Paper API
backend/app/api/papers.py - Ingestion pipeline
backend/app/ingestion/pipeline.py - arXiv client
backend/app/ingestion/arxiv_client.py - PDF parser
backend/app/ingestion/pdf_parser.py - Section chunker
backend/app/ingestion/chunker.py - Reference extraction
backend/app/ingestion/references.py - Paper summaries
backend/app/generation/summaries.py - Paper metadata
backend/app/models.py - SQLite metadata and runs
backend/app/db.py
Search and answers
- Search API
backend/app/api/search.py - Chat API
backend/app/api/chat.py - Query planning and RAG
backend/app/generation/rag.py - Hybrid retrieval
backend/app/retrieval/hybrid.py - Embedding models
backend/app/retrieval/embeddings.py - Cross-encoder reranker
backend/app/retrieval/reranker.py - Qdrant vector store
backend/app/retrieval/vector_store.py - Gemini client
backend/app/generation/llm.py - Citation validation
backend/app/generation/citations.py
Research tools
- Comparison API
backend/app/api/compare.py - Paper comparison
backend/app/generation/compare.py - Citation graph view
frontend/components/CitationGraph.tsx
Evaluation
- Evaluation API
backend/app/api/eval.py - Evaluation runner
backend/app/evaluation/runner.py - Evaluation dataset
backend/app/evaluation/dataset.py - Evaluation metrics
backend/app/evaluation/retrieval_metrics.py
Web interface
- Research UI
frontend/app/layout.tsx - Paper library
frontend/app/page.tsx - Search page
frontend/app/search/page.tsx - Chat page
frontend/app/chat/page.tsx - Compare page
frontend/app/compare/page.tsx - Evaluation page
frontend/app/eval/page.tsx
Other components
- Researcher
- FastAPI app
backend/app/main.py - Google Gemini
Connections 36
- Researcher → Research UI: uses
README.md - Research UI → Paper library: navigates
frontend/app/layout.tsx - Research UI → Search page: navigates
frontend/app/layout.tsx - Research UI → Chat page: navigates
frontend/app/layout.tsx - Research UI → Compare page: navigates
frontend/app/layout.tsx - Research UI → Evaluation page: navigates
frontend/app/layout.tsx - Paper library → Paper API: imports papers
README.md - Paper API → Ingestion pipeline: starts ingestion
backend/app/api/papers.py - Ingestion pipeline → arXiv client: fetches papers
backend/app/ingestion/pipeline.py - Ingestion pipeline → PDF parser: parses PDFs
backend/app/ingestion/pipeline.py - Ingestion pipeline → Section chunker: chunks sections
backend/app/ingestion/pipeline.py - Ingestion pipeline → Reference extraction: extracts references
backend/app/ingestion/pipeline.py - Ingestion pipeline → Paper summaries: generates summaries
backend/app/ingestion/pipeline.py - Ingestion pipeline → Qdrant vector store: upserts chunks
backend/app/ingestion/pipeline.py - Ingestion pipeline → Paper metadata: updates paper
backend/app/ingestion/pipeline.py - Paper API → SQLite metadata and runs: stores metadata
backend/app/api/papers.py - Search page → Search API: searches papers
README.md - Search API → Hybrid retrieval: runs search
backend/app/api/search.py - Chat page → Chat API: asks questions
README.md - Chat API → Query planning and RAG: requests answers
backend/app/api/chat.py - Query planning and RAG → Hybrid retrieval: retrieves passages
backend/app/generation/rag.py - Hybrid retrieval → Embedding models: embeds queries
backend/app/retrieval/hybrid.py - Hybrid retrieval → Qdrant vector store: queries vectors
backend/app/retrieval/hybrid.py - Hybrid retrieval → Cross-encoder reranker: optionally reranks
backend/app/retrieval/hybrid.py - Query planning and RAG → Gemini client: plans and generates
backend/app/generation/rag.py - Query planning and RAG → Citation validation: validates citations
backend/app/generation/rag.py - Gemini client → Google Gemini: calls model
README.md - Compare page → Comparison API: requests comparison
README.md - Comparison API → Paper comparison: builds comparison
backend/app/api/compare.py - Comparison API → SQLite metadata and runs: reads paper records
backend/app/api/compare.py - Researcher → Citation graph view: explores citations
README.md - Evaluation page → Evaluation API: runs evaluation
README.md - Evaluation API → Evaluation runner: starts evaluation
backend/app/api/eval.py - Evaluation API → Evaluation dataset: loads dataset
backend/app/api/eval.py - Evaluation runner → Evaluation metrics: computes retrieval metrics
README.md - Evaluation API → SQLite metadata and runs: stores run results
backend/app/api/eval.py