AI Autonomous Research Lab (ARL)

Multi-agent scientific research automation system built on Google's Agent Development Kit (ADK).

Overview

ARL automates the complete scientific research pipeline:

• Literature review and paper ingestion
• Hypothesis generation from research gaps
• Experiment design and validation
• Python code generation for experiments
• Sandboxed execution with result capture
• Statistical analysis and hypothesis validation
• Research report generation

Quick Start

CLI Tool

# Install uv (recommended)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install ARL
uv pip install -e .

# Set API key
export GOOGLE_API_KEY="your-key"

# Initialize
arl init

# Create project
arl project create --name "My Research" --domain cs

# Run research
arl research run --project <id> --request "Test hypothesis" --auto

Web UI (Backend + Frontend)

Terminal 1 - Backend:

cd arl-backend

# Start Docker services (PostgreSQL + Redis)
sudo systemctl start docker  # If not already running
docker compose up -d

# Install dependencies and run migrations
uv sync
.venv/bin/alembic upgrade head

# Start backend
.venv/bin/uvicorn app.main:app --reload

See arl-backend/SETUP.md for detailed setup instructions.

Terminal 2 - Frontend:

cd arl-frontend
npm install
echo "VITE_API_BASE_URL=http://localhost:8000" > .env
npm run dev

Access at http://localhost:5173

Features

• Multi-domain support: Computer Science, Biology, Physics, General Research
• Provider-agnostic LLM: Google Gemini, OpenAI, Anthropic, Azure OpenAI
• Hybrid deployment: Run locally or on Google Cloud
• A2A Protocol: Standardized agent-to-agent communication for microservices architecture
• Interactive collaboration: Human-in-the-loop at any stage
• Reproducible research: Complete experiment versioning and artifact management
• Docker sandbox: Secure isolated execution environment

Documentation

Getting Started

• Installation Guide - Setup and prerequisites
• Quickstart Guide - Get running in 5 minutes
• User Guide - Complete feature documentation

Architecture & Deployment

• Architecture Overview - System design and components
• A2A Protocol Guide - Agent-to-agent communication and microservices

Provider Setup

• Azure OpenAI Setup - Configure Azure OpenAI

Examples & References

• Examples Directory - Sample workflows and code
• Research Workflow Guide - Detailed workflow documentation

Architecture

Built on Google ADK with specialized agents:

• Orchestrator: Main workflow coordination
• Literature Agent: Paper ingestion and analysis
• Hypothesis Agent: Testable hypothesis generation
• Experiment Designer: Protocol and parameter specification
• Code Generator: Python code generation with validation
• Execution Engine: Docker sandbox execution
• Analysis Agent: Statistical validation and interpretation

See Architecture Overview for detailed system design.

Project Structure

arl/
├── docs/                      # Documentation
│   ├── installation.md
│   ├── quickstart.md
│   ├── user-guide.md
│   ├── architecture.md
│   ├── testing.md
│   ├── azure-setup.md
│   └── frontend-development.md
├── arl/                       # Main package
│   ├── adk_agents/            # ADK agent implementations
│   ├── core/                  # Core business logic
│   ├── integrations/          # External integrations
│   ├── cli/                   # Command-line interface
│   └── storage/               # Data persistence
├── arl-frontend/              # React web UI
├── tests/                     # Test suite
├── examples/                  # Example workflows
└── scripts/                   # Utility scripts

Technology Stack

• Framework: Google ADK 1.0+, Python 3.10+
• LLM: LiteLLM (Google Gemini, OpenAI, Anthropic, Azure OpenAI)
• Execution: Docker (sandboxed experiments)
• Storage: SQLite (local), Cloud SQL (cloud)
• Scientific Stack: NumPy, pandas, scikit-learn, PyTorch
• Frontend: React 18, TypeScript, Tailwind CSS, shadcn/ui

Contributing

We welcome contributions! Please see our development guides:

• Testing Guide
• Frontend Development

License

MIT License - See LICENSE file

Citation

If you use ARL in your research, please cite:

@software{arl2025,
  title={AI Autonomous Research Lab},
  author={ARL Team},
  year={2025},
  url={https://github.com/your-org/arl}
}