Budgetation

An AI-powered tool that analyzes bank statements out of PDFs, converts to CSVs, builds budget reports with spending type charts, and delivers a financial health score from deposits and withdrawals.

Comment

Budgetation: AI-Powered Personal Finance Assistant

Inspiration

Managing personal finances is often messy. PDFs from banks are hard to read, transactions are unorganized, and insights about spending are buried in raw data. We wanted to build an AI-powered finance assistant that automatically ingests statements, categorizes transactions, and produces insights that feel alive, adaptive, and personal. As college students who are struggling with budgeting and anxious about money, we find it absolutely necessary to have feedback on what we are doing.

What it does

  1. *Upload PDFs: Users upload their bank statements.
  2. Smart Ingestion (Attempted with Senso): The system extracts transactions, enriches them with Senso’s context layer, cleans the data, and stores it in a local database. This ensures both raw text and structured transaction data are captured for downstream insights.
  3. Categorization: Transactions are automatically categorized with a mix of rule-based logic (e.g. “Spotify goes into Entertainment”) and OpenAI for ambiguous cases.
  4. Dashboard Insights: A clean dashboard shows:
    • Net worth snapshot (income vs expenses)
    • Category-level pie chart spend analysis
    • Recurring subscription detection
    • Explain This Charge: Users can click on any transaction to get a contextual explanation, powered by Senso and OpenAI.
  5. Multilingual Reports: Attempted the summaries be instantly translated into other languages through DeepL

How we built it

  • Backend: Python
  • Frontend: Streamlit
  • AI Stack:
    • OpenAI for transaction categorization and reasoning.
    • Attempted Senso for contextual enrichment and rule-based triggers.
    • Attempted DeepL for language translation.
    • PDF Processing: PyPDF2 and regex-based parsing to extract transaction lines from messy bank statement layouts.

Challenges we ran into

  • Parsing PDFs: Different banks format statements differently, making regex extraction tricky. Some dates had no years, descriptions spanned multiple lines, and amounts appeared.

Accomplishments that we're proud of

  • Built a working MVP in under 5 hours that ingests PDFs, categorizes transactions, and produces insights.
  • Integrated multiple AI tools (OpenAI, Senso, DeepL) into one smooth workflow.
  • Designed a dashboard that’s both functional and visually clear.
  • Detected recurring subscriptions automatically from raw statement data.
  • Delivered a project that feels demo-ready and useful beyond a hackathon setting.

What we learned

  • Practical AI Integration: How to combine different AI APIs into one cohesive system.
  • Resilient Parsing: Handling messy real-world data (like unstructured PDFs) requires fallback strategies and flexible regex patterns.
  • Team Collaboration: Dividing tasks between ingestion, AI integration, and UI was crucial to finish on time.
  • User-Centric Thinking: Insights need to be actionable (not just pretty charts) for users to actually benefit.

What's next for Budgetation

  • Predictive Analytics: Show forecasts like “Your subscription spend will exceed $400 next month.”
  • Mobile App: Build a mobile-first UI for on-the-go finance tracking.
  • Budget Coaching: Personalized AI-driven recommendations for saving, investing, and debt reduction.
  • Security & Privacy: Implement encryption, secure auth, and user data isolation for a finance app.
  • Company-based: Through an implementation of Clickhouse, more comprehensive and huge reports can be done with much more transactions and bigger amounts.

Built With

  • clickhouse(future)
  • deepl(attempted)
  • openai
  • soro(attempted)
Share this project:

Updates