FinSight AI

• 

  AI agent answering financial questions in natural language using MongoDB Atlas data

• 

  Real-time spending dashboard with pie chart and bar chart powered by MongoDB Atlas

• 

  AI-powered anomaly detection — budget overruns and forgotten subscriptions detected automatically

• 

  Complete transaction history stored and queried from MongoDB Atlas in real time

• 

  Users can add new transactions live — the AI agent instantly reasons over updated MongoDB data in real time.

💡 Inspiration

Most people have no idea where their money goes. Traditional finance apps show numbers but don't reason over them. I wanted to build an agent that doesn't just display data — it thinks, plans, and tells you what matters.

🤖 What it does

FinSight AI is a personal finance agent that connects to MongoDB Atlas via MCP and gives you a financial advisor in your pocket:

• Natural language queries — Ask "How much did I spend on food?" and get instant analysis
• Anomaly detection — Automatically detects budget overruns and forgotten subscriptions
• Spending forecast — Projects your end-of-month savings based on current trends
• Multi-step reasoning — The agent plans, queries MongoDB, analyzes, and responds
• Live dashboard — Real-time pie charts and bar charts powered by MongoDB Atlas data
• Transaction history — 84+ transactions across 3 months with categories and amounts
• Add transactions — Users can add new transactions live and the agent immediately reasons over them
• Budget tracking — 7 budget categories with real-time overspending alerts

🏗️ How I built it

Backend: FastAPI + Python running on Render AI Brain: Groq LLaMA 3.3 70B for fast, intelligent reasoning Database: MongoDB Atlas as the financial data backbone MCP Server: Custom MongoDB MCP server exposing 5 tools to the agent Frontend: React + Vite + Tailwind CSS + Recharts Deployment: Render (backend) + Netlify (frontend)

The MongoDB MCP server exposes 5 tools:

1. get_transactions — Fetch all transactions
2. get_spending_by_category — Aggregate spending
3. get_budget_status — Compare actual vs budget
4. add_transaction — Write new transactions
5. detect_anomalies — Detect budget overruns

🚧 Challenges I faced

• Migrating from deprecated google-generativeai to the new google-genai SDK
• Handling MongoDB connection strings with special characters in URLs
• Getting CORS to work between Netlify frontend and Render backend
• Making the agent responses feel natural and actionable, not robotic

📚 What I learned

• How to build and expose MCP servers for AI agents
• How MongoDB Atlas aggregation pipelines power real-time financial analytics
• How to architect a full-stack AI agent with FastAPI + React
• The power of LLaMA 70B for complex multi-step reasoning tasks

🏆 Accomplishments

• Built a fully functional AI finance agent with 5 features
• 84+ transactions across 3 months stored in MongoDB Atlas
• Live Add Transaction feature — agent updates in real time
• Deployed live with Render + Netlify — fully accessible to judges
• MCP server gives the agent true database superpowers
• The agent detects anomalies and forecasts spending automatically

Built With

• atlas
• css
• fastapi
• groq
• llama
• mongodb
• netlify
• python
• react
• render
• tailwind