Self-Correcting RAG for Scientific Research
A retrieval-augmented generation system designed for scientific literature. NexusRAG reduces hallucinations through a self-correcting pipeline, provides precise citations (exact passages and page numbers), runs entirely locally via Ollama, and synthesizes insights across multiple documents.
You upload research papers (PDF, DOCX, TXT, MD). NexusRAG parses and chunks them, stores embeddings in LanceDB, and builds a BM25 keyword index alongside it. When you ask a question, a hybrid retriever combines semantic search with keyword matching to find the most relevant passages. Those passages go to a local LLM (via Ollama) which generates an answer with inline citations. Before returning, a verification step checks each citation against the source — if something doesn't hold up, the system reformulates the query, re-retrieves, and tries again. The result is an answer grounded in your documents with traceable references.
- Hallucination reduction — Self-correcting retrieval loop verifies claims against source documents
- Precise citations — Every claim links to exact passages and page numbers
- Local-only processing — All data stays on your machine via Ollama (no API keys needed)
- Cross-document synthesis — Answers draw from and connect insights across multiple papers
# Clone and install
git clone https://github.com/urme-b/NexusRAG.git
cd NexusRAG
python -m venv .venv && source .venv/bin/activate
pip install -e .
# Start Ollama (separate terminal)
ollama serve
ollama pull llama3.2:3b
# Launch NexusRAG
make run
# → http://localhost:8000| Model | Size | RAM Required | Speed | Best For |
|---|---|---|---|---|
llama3.2:3b |
2 GB | 8 GB | Fast | Quick queries, basic summarization |
llama3.1:8b |
4.9 GB | 16 GB | Moderate | Complex multi-document analysis |
Switch models by updating the configuration — see configs/ for details.
Python · FastAPI · LanceDB · Ollama · Sentence Transformers
- Python 3.11+
- 8 GB RAM minimum (16 GB recommended for larger models)
- Ollama installed and running
- Latency reduction for real-time querying
- Multimodal support (tables, figures, charts from PDFs)
- Collaborative knowledge base for research teams
- Streaming responses in the web interface
- Support for additional LLM providers beyond Ollama
- Batch ingestion with progress tracking
- Export citations in standard formats (BibTeX, APA)
This project is licensed under the MIT License.