Ledger

Ledger combines bank data, email receipts, and AI agents to automatically track spending, detect subscriptions, forecast cashflow, and help users make smarter financial decisions.

Comment

Inspiration

Personal finance today is fragmented. Your bank shows transactions, your email contains receipts and subscriptions, and budgeting apps rely on manual input or static rules. This makes it difficult for people to understand where their money is going or make smarter financial decisions.

We wanted to build something that acts less like a spreadsheet and more like a financial copilot. Ledger brings together bank data, email receipts, and AI-powered agents to give users a clear, intelligent view of their finances and actionable insights about their spending.

What it does

Ledger is an AI-powered personal finance platform that automatically tracks spending, detects subscriptions, forecasts cashflow, and helps users plan financial goals.

The platform connects to:

  • Bank accounts via Plaid
  • Email inboxes via Gmail OAuth to scan receipts and billing emails

Ledger analyzes the data using a team of specialized AI agents:

  • Pulse — cashflow forecasting and balance prediction
  • Audit — subscription detection and expense analysis
  • North Star — savings goal planning and feasibility analysis
  • Sentinel — anomaly detection and spending alerts

These agents are coordinated by a Council orchestrator, which routes user questions to the appropriate specialist and synthesizes responses.

Users can explore their finances through:

  • A cashflow dashboard with historical and projected views
  • Automatic subscription detection from transactions and email receipts
  • Savings goal planning with feasibility analysis
  • A chat interface that answers financial questions using real data

How we built it

Ledger is a full-stack application built using modern web technologies and AI infrastructure.

Frontend

  • React 18
  • TypeScript
  • Vite
  • Tailwind CSS

Backend

  • FastAPI (Python)
  • Pydantic for validation
  • Auth0 for authentication

Data & Integrations

  • Supabase (PostgreSQL + Row Level Security)
  • Plaid API for bank account integration
  • Gmail API for receipt and billing email scanning
  • Backboard.io for multi-agent architecture
  • Railway for deployment

AI System

Ledger uses a multi-agent architecture where specialized agents analyze financial data independently while an orchestrator coordinates their responses.

Each agent focuses on a specific financial domain while remaining stateless, ensuring consistent and explainable outputs.

Financial Forecasting

Ledger estimates future balances based on income, recurring expenses, and projected spending.

For example, the projected balance over time can be approximated as:

$$ B_{t+1} = B_t + I_t - E_t $$

Where:

  • (B_t) = current balance
  • (I_t) = income during period (t)
  • (E_t) = expenses during period (t)

This allows the system to forecast potential low-balance events and recommend timing strategies for spending or saving.

Challenges we ran into

One of the biggest challenges was coordinating multiple AI agents in a reliable way. We needed a system that could decide which agent should handle a user request while still producing a clear, unified response.

Another challenge was detecting recurring subscriptions from inconsistent data sources. Transaction names and email receipts often vary significantly, so we combined pattern detection with LLM-powered extraction.

We also had to carefully handle financial integrations such as Plaid and Gmail OAuth while ensuring secure authentication and data synchronization.

What we learned

Building Ledger taught us a lot about:

  • Designing multi-agent AI systems
  • Combining API-driven data pipelines with LLM analysis
  • Building scalable full-stack apps with React and FastAPI
  • Working with real-world financial data, which is often messy and inconsistent

We also learned that users don’t just want dashboards, they want actionable insights about their money.

What's next for Ledger

Future improvements could include:

  • Automated budget optimization
  • Smart subscription cancellation suggestions
  • Real-time financial alerts
  • Additional financial integrations
  • Mobile support and push notifications

Our long-term goal is to turn Ledger into a true AI financial copilot that helps users understand and improve their financial health. An immediate goal for the future is to actually have the app version published instead of having a web app.

Built With

Share this project:

Updates