Financial Advisor AI Assistant

A production-ready AI assistant for financial advisors that integrates with Gmail, Google Calendar, and HubSpot CRM. The assistant provides intelligent responses using RAG (Retrieval-Augmented Generation) and can perform actions through tool calling.

Features

• Authentication: Google OAuth (Gmail + Calendar) and HubSpot OAuth
• RAG Pipeline: Vector embeddings of emails and CRM data for intelligent responses
• Tool Calling: Automated actions across Gmail, Calendar, and HubSpot
• Proactive Agent: Memory-based instructions and event-driven actions
• Chat Interface: ChatGPT-like streaming interface with responsive design

Tech Stack

• Frontend: React + TailwindCSS
• Backend: Python + FastAPI
• Database: PostgreSQL + pgvector
• AI: OpenAI API with embeddings and tool calling
• Auth: OAuth 2.0 (Google, HubSpot)
• Deployment: Docker + Docker Compose

Project Structure

advisor-ai/
├── frontend/          # React frontend
├── backend/           # FastAPI backend
├── database/          # Database migrations and schemas
├── docker-compose.yml # Local development setup
├── Dockerfile         # Production Docker configuration
└── README.md          # This file

Quick Start

Prerequisites

• Docker and Docker Compose
• Google Cloud Console project with OAuth credentials
• HubSpot developer account
• OpenAI API key

Environment Variables

Create .env files in both frontend/ and backend/ directories:

Backend .env:

DATABASE_URL=postgresql://user:password@localhost:5432/advisor_ai
OPENAI_API_KEY=your_openai_api_key
GOOGLE_CLIENT_ID=your_google_client_id
GOOGLE_CLIENT_SECRET=your_google_client_secret
HUBSPOT_CLIENT_ID=your_hubspot_client_id
HUBSPOT_CLIENT_SECRET=your_hubspot_client_secret
SECRET_KEY=your_secret_key_for_jwt

Frontend .env:

REACT_APP_API_URL=http://localhost:8000
REACT_APP_GOOGLE_CLIENT_ID=your_google_client_id
REACT_APP_HUBSPOT_CLIENT_ID=your_hubspot_client_id

Development Setup

1. Clone the repository
2. Set up environment variables
3. Start the development environment:

docker-compose up -d

4. Access the application:
   • Frontend: http://localhost:3000
   • Backend API: http://localhost:8000
   • API Documentation: http://localhost:8000/docs

Production Deployment

The application is configured for deployment on Render or Fly.io with Docker support.

Architecture

RAG Pipeline

1. Data Ingestion: Gmail emails + HubSpot contacts/notes → embeddings → pgvector
2. Query Processing: User question → embedding → similarity search → context retrieval
3. Response Generation: Context + question → LLM → answer or action JSON

Tool Calling System

1. Action Parsing: LLM outputs structured JSON
2. Validation: Check contact existence, time availability, etc.
3. Execution: Call Gmail, Calendar, HubSpot APIs
4. Memory: Store tasks with status for continuation

Proactive Agent

• Memory System: Persistent storage of ongoing instructions
• Event Triggers: Webhooks/polling from Gmail, Calendar, HubSpot
• Action Execution: Automatic responses to events based on instructions

API Endpoints

Authentication

• POST /auth/google - Google OAuth callback
• POST /auth/hubspot - HubSpot OAuth callback
• GET /auth/me - Get current user info

Chat

• POST /chat/message - Send message to AI assistant
• GET /chat/history - Get chat history
• POST /chat/stream - Streaming chat endpoint

Actions

• POST /actions/execute - Execute tool calling action
• GET /actions/tasks - Get pending tasks
• PUT /actions/tasks/{task_id} - Update task status

RAG

• POST /rag/ingest - Ingest new data (emails, contacts)
• GET /rag/search - Search vector database
• POST /rag/query - Query with context retrieval

Development Standards

• Code Quality: TypeScript for frontend, Python + Pydantic for backend
• Error Handling: Graceful error handling with structured logging
• Testing: Unit and integration tests where appropriate
• Security: OAuth 2.0, JWT tokens, secure credential storage
• Performance: Async/await, connection pooling, efficient queries

Contributing

1. Follow the established code style and patterns
2. Write tests for new features
3. Update documentation as needed
4. Ensure all CI/CD checks pass

License

Private - All rights reserved