Deep Research

Overview

A multi-agent AI research assistant, powered by Gemini 2.0 Flash, built with Google ADK that creates targeted research plans and automatically researches and answers your questions with comprehensive, well-cited responses.

Workflow

The system processes user queries through four main phases:

1. Classification - Determines if the request is a valid research question, needs more information, or is a general query
2. Planning - Creates a targeted research plan with up to 3 steps, combining ArXiv searches for academic foundations and web searches for current developments
3. Research Execution - Runs the plan in parallel, with specialized agents handling:
   • ArXiv Research: Scans top 50 papers, selects top 3 based on abstracts, ingests them into Qdrant with OpenAI embeddings, and performs RAG to extract relevant information while preserving URLs for citations
   • Web Search: Uses Tavily MCP server to gather current information and developments
4. Answer Synthesis - Aggregates all findings and generates a comprehensive, well-cited response using the collected research data

Getting Started

Prerequisites

• Python 3.11+
• API keys for supported AI providers:
  • Google AI API key (for Gemini models)
  • OpenAI API key (for embeddings, Gemini embeddings not yet implemented)
  • Tavily API key (for web search)
• Docker (for local Qdrant setup) or access to a cloud vector database

Setup

1. Clone the repository:

   git clone <repo-url>
   cd deep-research

2. Install dependencies:

   uv sync

3. Configure environment variables:

   Copy the example environment file:

   cp multi_tool_agent/.env.example multi_tool_agent/.env

   Edit the .env file to set your API keys:

   # Google AI Configuration
   GOOGLE_API_KEY=your_google_api_key_here
   
   # OpenAI and Search API Configuration
   OPENAI_API_KEY=your_openai_api_key_here
   TAVILY_API_KEY=your_tavily_api_key_here
   
   # Vector Store Configuration (default: Qdrant)
   VECTOR_STORE_TYPE=qdrant
   QDRANT_HOST=localhost
   QDRANT_PORT=6333
   QDRANT_GRPC_PORT=6334
   QDRANT_PREFER_GRPC=true
   
   # Logging Configuration
   LOG_LEVEL=INFO

4. Set up vector database:

   For local development with Docker:

   docker run -p 6333:6333 -p 6334:6334 qdrant/qdrant

   Alternatively, you can use a cloud provider (Qdrant Cloud, etc.) or implement your own vector database by extending the base classes in multi_tool_agent/data/vector_stores/.

5. Execute the research system:

   uv run adk run multi_tool_agent

Usage Examples

Basic Research Query

• Who is the current president of the USA?
• Tell me the latest news on Elon Musk
• What are the latest developments in transformer architecture optimization?

Domain-Specific Research

• How do quantum computing algorithms compare to classical machine learning for optimization problems?
• What are the current applications of CRISPR technology in treating genetic disorders?
• What are the latest developments in renewable energy storage solutions?

demo.mp4