Local AI Research Engine

Local AI Research Engine is an advanced RAG system that runs 100% locally on your machine. By combining semantic search with knowledge graphs, it helps researchers analyze complex document libraries without compromising privacy. Features include automated paper comparisons, contradiction detection, and publication-ready literature reviews

Features

• Document Intelligence: Ingest PDFs, markdown, code, and text files
• Knowledge Graph: Automatically extract entities and relationships
• Hybrid Retrieval: Vector search + keyword search + graph traversal
• Cited Answers: Every answer includes source citations
• Multi-Document Reasoning: Synthesize information across multiple sources
• Advanced Analysis: Paper comparisons, literature reviews, and contradiction detection
• 100% Local: Runs entirely on Ollama - no API calls, complete privacy

Prerequisites

1. Python 3.9+
2. Ollama installed and running
3. Required Models:

   ollama pull mistral:latest
   ollama pull nomic-embed-text

Installation

1. Clone or download this repository
2. Create a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Copy .env.example to .env and configure if needed

Quick Start

1. Add documents to data/documents/
2. Run the Streamlit UI:

   streamlit run ui/streamlit_app.py

3. Or use the CLI:

   python main.py

Project Structure

local-research-engine/
├── ingest/          # Document loading and chunking
├── index/           # Vector store, keyword index, knowledge graph
├── retrieval/       # Hybrid search and reranking
├── llm/             # Ollama client and prompts
├── ui/              # Streamlit interface
├── data/            # Documents and indexes
└── main.py          # CLI interface

Usage

Upload Documents

Place your documents in data/documents/ or use the Streamlit upload interface.

Ask Questions

The system will:

1. Retrieve relevant chunks using hybrid search
2. Expand context using the knowledge graph
3. Rerank evidence with LLM
4. Generate an answer with inline citations

Example Query

Q: "What is the EM algorithm used for?"

A: The EM algorithm is used to estimate HMM parameters by iteratively maximizing the expected log-likelihood [Rabiner1989.pdf §4]. Unlike gradient-based methods, EM guarantees non-decreasing likelihood [Bishop.pdf §9.2].

Configuration

Edit config.yaml to customize:

• Chunk sizes
• Retrieval parameters
• Model selection
• Storage paths

License

MIT License - See LICENSE file for details