FinQuery Financial Agentic System

An AI-powered financial advisor that helps users find the best savings accounts, ISAs, loans, and credit cards using real-time banking data.

Comment

Inspiration

I wanted to build something useful for myself. Sometimes, I don’t want to spend time checking the latest bank products. Traditional comparison websites are fine, but with the rise of AI, I figured we could speed things up. Instead of manually searching, why not just chat with an AI? You can ask about the latest products from banks and get financial advice along the way.

What it does

Right now, it analyses a dataset I created using Faker in Python. This dataset includes fake bank data and product details. Based on the user’s question, the app generates an AQL query to pull relevant information and summarises it in a way that answers the question. It also looks at past chats to see if any of the previous context is helpful for the current query, making responses more personalised.

How I built it

  • Frontend: I used React.js for the user interface.
  • Backend: The backend is built with Python and FastAPI. Flask could’ve worked too, but FastAPI is faster for building API routes and offers automatic documentation.
  • Database: I used ArangoDB to store both the dataset and chat messages. It’s great for handling complex data and queries.
  • AI Integration: The app connects to the Groq API to handle AI-powered responses.

Challenges we ran into

  • Groq Limitations: Groq has a character limit per request, so analysing long chat histories with the free tier can be tricky.
  • Designing the Agentic System: I’m not completely sure if the way I split up tasks between agents is the best method. There are many different ways to approach this, and I think the system could be more efficient with more time.

Accomplishments that we're proud of

  • Building a Full-Stack App: I’m proud of creating a full-stack app on my own in such a short amount of time.
  • Creating the Agentic System: I built the AI system from scratch with very little help—an amazing learning experience!
  • Inspiring Others: I hope other developers can learn from this and take it even further.

What I learned

  • Prompt Engineering: I got better at writing prompts that make AI give useful responses.
  • Task Breakdown: I improved at breaking down large problems into smaller tasks and letting different agents handle them.
  • Frontend and Backend Integration: Combining the frontend with the backend was a challenge, but it was a great learning experience.

What's next for AI Personal Financial Advisor & Graph Intelligence

  • Better Chat History Analysis: I can improve how the app uses past chats to make responses more intelligent.
  • More Agents: Adding more agents will help break down tasks further, making the system even smarter.
  • Improve User Experience: I want to make the app sound less robotic and more natural when responding to users.
  • Better Dataset: Right now, I’m using a fake dataset because I couldn’t find a real one that fits. If I can find a more accurate, up-to-date dataset—like the ones used by real comparison sites—it would make the app much more reliable.
  • Cover More User Inputs: I want to expand the range of user inputs the system can handle, so it can respond to a wider variety of questions and use cases.

Built With

Share this project:

Updates