Research-Agent

# Research Agent

A deep research AI system that generates publication-ready LaTeX papers. Built with a self-hosted reasoning model (DeepSeek-R1-Distill-Qwen-14B) and GPT-4o-mini workers for search and extraction.

## Features

- **Deep Research**: Searches web and ArXiv for relevant papers and sources
- **Self-Hosted Brain**: 128K context reasoning model via vLLM
- **Publication-Ready Output**: LaTeX papers with proper BibTeX citations
- **Domain Focus**: AI/ML, Quantum Physics, Astrophysics
- **Resilient**: Circuit breaker protection for external services
- **Observable**: Prometheus-compatible metrics and structured logging

## Architecture

```
User Query
    |
    v
+------------------+
|   FastAPI        |  API Layer (rate limiting, validation)
+------------------+
    |
    v
+------------------+
|   LangGraph      |  Orchestration (state machine, checkpointing)
+------------------+
    |
    +---> Brain (DeepSeek-R1-Distill-14B, self-hosted vLLM)
    |       - Planning: Creates research strategy
    |       - Analysis: Evaluates gathered sources
    |       - Synthesis: Combines findings
    |       - Review: Quality checks output
    |
    +---> Workers (GPT-4o-mini API)
            - Web search (Tavily API)
            - ArXiv search and paper retrieval
            - LaTeX section drafting
    |
    v
+------------------+
|   LaTeX Output   |  Papers with BibTeX citations
+------------------+
```

## Requirements

- Python 3.12+
- Docker and Docker Compose
- NVIDIA GPU with 24GB+ VRAM (for local brain service)
- API keys: OpenAI, Tavily

## Quick Start

### 1. Clone and setup

```bash
git clone <repository-url>
cd research-agent
./scripts/setup.sh
```

### 2. Configure environment

```bash
cp .env.example .env
```

Edit `.env` with your API keys:

```bash
# Required
OPENAI_API_KEY=sk-your-openai-api-key
TAVILY_API_KEY=tvly-your-tavily-api-key

# Brain service (use mock for development)
BRAIN_SERVICE_URL=http://localhost:8001
BRAIN_API_KEY=your-internal-bra