The Problem
59M US freelancers. 73% don't systematically track deductions. Average missed: ~$2,400/year. That's $141B left on the table nationally.
What I Built
Connect your bank (via Plaid) → AI classifies every transaction against IRS Schedule C rules → shows you exactly what you're missing, broken down by category.
Here's what it finds in a typical freelancer's transactions:
| Transaction | Amount | AI Classification | You Save |
|---|---|---|---|
| Zoom Pro Monthly | $14.99 | Business — Software | $14.99 |
| WeWork Hot Desk | $350.00 | Business — Rent | $350.00 |
| Uber to client meeting | $24.50 | Business — Travel | $24.50 |
| Comcast Internet | $89.00 | Mixed — 60% Business | $53.40 |
| Netflix Premium | $15.99 | Personal | $0 |
| Starbucks w/ client | $12.80 | Business — Meals 50% | $6.40 |
Stack
- Frontend: React + TypeScript + Tailwind + Vite
- Backend: Express + PostgreSQL (event sourced)
- Bank data: Plaid API
- AI classification: Claude API (two-stage pipeline — extract 5W1H events → classify against IRS rules)
- MCP server: 4 tools for headless tax classification
- Learning system: Patterns improve with every user correction
The AI uses a confidence scoring system — high-confidence deductions are auto-classified, edge cases get flagged for your review. You always have final say.
The Ask
Looking for freelancers/contractors to test the early version. Free forever for early testers. I want to make sure it works for different types of freelance work before launching properly.
Interested? Sign up for early access: https://chrbailey.github.io/deductai/
Would love feedback on the landing page too — does the value prop land?
Built as a Stanford TECH42 project. Open to questions about the architecture or approach.
Top comments (0)