Financial Roaster

A small AWS Bedrock app that reviews spending data with four parallel agents and returns a roast-style financial summary.

Try It Out

Launch Financial Roaster

Overview

Financial Roaster reviews spending data and turns it into a roast-style summary. Four agents handle cleanup, pattern detection, risk checks, and the final write-up.

Tech Stack

  • AI Engine: AWS Bedrock + Claude 3 Haiku
  • Workflow: LangGraph + Agent Core
  • Backend: FastAPI (Python)
  • Frontend: HTML/CSS/JavaScript
  • Deployment: Render
  • OCR: PyMuPDF + EasyOCR

Multi-Agent Architecture

The app uses four agents:

1. The Accountant

  • Cleans and normalizes transaction names
  • Categorizes uncategorized transactions
  • Flags data quality issues

2. The Detective

  • Identifies recurring subscriptions
  • Detects spending patterns and trends
  • Highlights frequent merchants

3. The Risk Manager

  • Identifies potential overdraft behavior
  • Flags impulse purchase patterns
  • Detects concerning spending habits

4. The Judge

  • Calculates an overall financial health score (0-100)
  • Synthesizes findings from other agents
  • Generates the final analysis

Parallel Processing

Agents run concurrently using Python’s ThreadPoolExecutor, reducing analysis time by approximately 50%:

with ThreadPoolExecutor(max_workers=3) as executor:
    detective_future = executor.submit(find_patterns)
    risk_manager_future = executor.submit(assess_risks)
    judge_future = executor.submit(calculate_chaos)

Key Features

  • Multi-format input: Supports CSV, PDF, images, and manual text entry
  • Smart parsing: Combines regex and AI for accurate data extraction
  • Interactive results: 3D flip cards for shareable output
  • Mobile responsive: Works on phone and desktop screens
  • Cost efficient: ~$0.01 per analysis

Deployment

  • Platform: Render (auto-deploys from GitHub)
  • Environment: Python 3.11 + AWS Bedrock
  • Caching: File-hash based caching to avoid redundant processing