ExpenseAI

• 

  a dashboard after user did the query input.

📖 Project Story

🚀 About the Project

The Agentic AI Expense Management System is an intelligent financial assistant designed to help users understand, analyze, and optimize their spending habits. Instead of just tracking expenses, the system uses a multi-agent AI architecture to simulate decision-making processes similar to human financial reasoning—classifying expenses, detecting patterns, assessing risk, and generating personalized suggestions.

---

💡 Inspiration

The idea came from a common problem:
Most expense trackers only record data but don’t provide meaningful insights. Users still struggle to answer questions like:

• Am I overspending?
• Where can I cut costs?
• What habits are hurting my finances?

We wanted to build something smarter—an AI-powered system that doesn’t just track money, but thinks about it.

---

🏗️ How We Built It

The project was built using a full-stack architecture:

• Backend: FastAPI for high-performance APIs
  
• Database: MongoDB for flexible data storage
  
• Authentication: JWT + bcrypt for secure login
  
• AI Layer: Google Gemini API + LangChain-inspired agent flow
  
• Frontend: Streamlit dashboard for an interactive user experience
  

🤖 Multi-Agent System Design

We designed the system as a pipeline of specialized AI agents:

• Ingestion Agent → Processes raw expense data
  
• Classification Agent → Categorizes expenses (essential/non-essential)
  
• Pattern Agent → Identifies spending trends
  
• Risk Agent → Evaluates financial risk
  
• Suggestion Agent → Generates actionable insights
  

All agents are orchestrated through a central Orchestrator, creating a modular and scalable AI workflow.

---

🧠 What We Learned

• How to design and implement agentic AI systems
• Integrating LLMs (Gemini API) into real-world applications
  
• Building secure authentication using JWT and hashing
  
• Handling async database operations with MongoDB
  
• Creating interactive dashboards with Streamlit & Plotly
  
• Debugging real-world issues like API mismatches, environment configs, and dependency conflicts
  

---

⚔️ Challenges We Faced

1. 🔐 Authentication Flow Mismatch

Frontend and backend initially had mismatched response formats (e.g., token vs access_token), which caused login failures.

2. 🧩 Multi-Agent Coordination

Designing a clean flow between agents without tightly coupling them required careful structuring.

3. ⚙️ Environment & Dependency Issues

• Virtual environment path errors
  
• Package build failures (like hnswlib)
  
• MongoDB connection authentication issues
  

4. 🌐 Frontend–Backend Integration

Ensuring smooth communication between Streamlit and FastAPI APIs required debugging payload formats and response handling.

---

🎯 Outcome

The final system provides:

• 📊 Real-time expense analysis
  
• ⚠️ Risk detection and alerts
  
• 💡 AI-generated financial suggestions
  
• 📈 Interactive dashboards for better visualization
  

---

🔮 Future Improvements

• Real-time expense syncing
  
• Mobile app integration
  
• Advanced AI insights using embeddings and vector databases
  
• Personalized financial goal tracking
  
• Deployment on cloud (AWS / Render / Docker)

---

❤️ Final Thoughts

This project demonstrates how Agentic AI can move beyond chatbots and become a powerful decision-support system. It’s not just about tracking expenses—it’s about understanding and improving financial behavior through AI.

Built With

• fastapi
• google-gemini-api
• jwt-auth
• langchain
• mongodb
• python