Presidio Research Agent with A2A Protocol

This project implements an Internal Research Agent for Presidio Insurance that uses LangChain, Google Docs MCP integration, and Google's A2A (Agent-to-Agent) protocol to deliver accurate, contextual, and actionable responses to employee queries.

Features

• Google Docs Integration: Search and analyze documents stored in Google Drive
• RAG System: Search and retrieve answers from HR Policy documents
• Web Search: Fetch industry benchmarks, trends, and regulatory updates
• A2A Protocol: Agent-to-agent communication for complex tasks
• Advanced Document Analysis: Document summarization, table extraction, and trend analysis

Architecture

The system follows a modular architecture:

User → React Frontend → FastAPI Backend → LangChain Agent → Tools (MCP, RAG, Web Search)

Components

1. Frontend: React-based UI for user interaction
2. Backend: FastAPI server that handles requests and manages services
3. Agent Service: LangChain-based agent that orchestrates tools
4. MCP Server: Custom MCP server for Google Docs integration
5. A2A Service: Implementation of Google's A2A protocol

Setup

Prerequisites

• Python 3.8+
• Node.js 14+
• Google Cloud Platform account with API access
• SerpAPI key for web search

Environment Variables

Create a .env file with the following variables:

# Google API
GOOGLE_APPLICATION_CREDENTIALS=credentials.json

# SerpAPI
SERPAPI_API_KEY=your_serpapi_key

# AWS (for Bedrock)
AWS_ACCESS_KEY_ID=your_aws_access_key
AWS_SECRET_ACCESS_KEY=your_aws_secret_key
AWS_REGION=us-east-1

# Vector Database
PINECONE_API_KEY=your_pinecone_key
PINECONE_INDEX_NAME=presidio-research

Installation

1. Install Python dependencies:

   pip install -r requirements.txt
   

2. Install frontend dependencies:

   cd src/frontend
   npm install
   

3. Set up Google OAuth:

   python run_mcp_local.py
   

4. Start the application:

   ./run.sh
   

Usage

Example Queries

The agent can handle various types of queries:

1. Document Research:

   • "Summarize all customer feedback related to our Q1 marketing campaigns."
   • "Extract key insights from our customer satisfaction surveys."

2. Policy Compliance:

   • "Find relevant compliance policies related to AI data handling."
   • "What are our guidelines for remote work?"

3. Market Intelligence:

   • "Compare our current hiring trend with industry benchmarks."
   • "What are the latest regulatory updates in insurance?"

A2A Protocol

The system uses Google's A2A protocol for agent-to-agent communication. This allows specialized agents to collaborate on complex tasks:

1. Document Research Agent: Specializes in searching and analyzing Google Docs
2. Policy Compliance Agent: Focuses on HR policies and compliance requirements
3. Market Intelligence Agent: Gathers industry benchmarks, trends, and regulatory updates

API Endpoints

Chat API

• POST /api/chat: Process a chat message using the LangChain agent

A2A Protocol API

• POST /api/a2a/send_message: Send a message from one agent to another
• GET /api/a2a/conversation/{conversation_id}: Get the history of an A2A conversation
• POST /api/a2a/process_query: Process a user query using the A2A protocol

Document API

• GET /api/documents: List all documents
• POST /api/documents/upload: Upload and process a document
• GET /api/documents/{document_id}: Get document by ID
• DELETE /api/documents/{document_id}: Delete a document
• POST /api/documents/search: Search for documents based on query

Tool API

• POST /api/tools/execute: Execute a specific agent tool directly

MCP Server Endpoints

• /search_google_docs: Search Google Docs for information
• /extract_document_content: Extract the full content of a Google Doc
• /analyze_document_trends: Analyze trends in time-series data within Google Sheets
• /compare_document_trends: Compare trends across multiple Google Sheets
• /summarize_document: Generate a summary of a Google Doc
• /extract_document_tables: Extract tables from a Google Doc

Development

Adding New Tools

To add a new tool:

1. Implement the tool function in the appropriate service
2. Add the tool to the _get_tools method in agent_service.py
3. Update the system message to include the new tool

Adding New Agents

To add a new specialized agent:

1. Create a message handler function in agent_service.py
2. Register the agent in the _initialize_specialized_agents method
3. Add an endpoint in the MCP server to handle requests for this agent

References

• LangChain Documentation
• Google's A2A Protocol
• Creating MCP Server - YouTube