Inspiration
- Freelancers lose money in two places: underpricing their work and chasing late or disputed payments.
- Most invoices are priced by gut feeling, and disputes turn into long, awkward email threads.
- We wanted an invoice that is defensible before it is sent, with PayPal as the backbone for getting paid.
What it does
- Reads the client contract and the freelancer's work logs.
- Extracts rates, scope, deliverables and payment terms.
- Scores the invoice for fairness against market rates and work complexity.
- Drafts line items with a plain-language justification for each.
- Sends the invoice through PayPal Invoicing, and the client pays with a PayPal button.
- Tracks disputes through webhooks and drafts evidence-backed replies.
Fairness score
$$ F = w_1 \cdot \frac{r_{\text{billed}}}{r_{\text{market}}} + w_2 \cdot \frac{h_{\text{billed}}}{h_{\text{expected}}}, \qquad w_1 + w_2 = 1 $$
- F close to 1 means the invoice is in line with the market.
- F well above 1 is flagged for review before sending.
- F well below 1 tells the freelancer they are undercharging.
How we built it
- AI layer: Claude/GPT API for contract parsing, rate benchmarking, invoice drafting and dispute replies.
- Structured JSON outputs so every number can be validated before it reaches a client.
- PayPal Invoicing API to create and send invoices.
- PayPal Orders API for the pay-now checkout flow.
- PayPal webhooks to track payment status and disputes, which update the dashboard and trigger the AI dispute drafter.
- Frontend and backend: add your stack here (for example Next.js, FastAPI, Postgres).
Challenges we ran into
- Hallucinated market rates: fixed by making the model cite its benchmark source and range, and clamping outputs to validated limits.
- Messy inputs: contracts and work logs come in inconsistent formats and needed normalizing first.
- Webhook reliability: handled retries, duplicate events and signature verification so payment status never went out of sync.
- Dispute tone: replies had to be firm but professional, and a human approves them before sending.
Accomplishments that we're proud of
- A full loop from contract upload to a paid invoice, with PayPal at the center.
- A fairness score that explains why, not just a number.
- Dispute drafts that cite the exact contract clause and work log entries.
What we learned
- Designing AI outputs as verifiable structured data instead of free text.
- The PayPal Invoicing lifecycle and webhook event model.
- Trust is the product: both freelancer and client need to see the reasoning.
What's next
- Live rate data from multiple market sources.
- Time tracker integrations (Toggl, Harvest) to pull work logs automatically.
- Recurring invoices and milestone-based billing.
- Automatic late-payment reminders.
Built With
- and
- anthropic-api
- backend
- claude
- data
- fastapi
- frontend
- json
- nextjs
- openai
- paypal
- paypal-invoicing-api
- paypal-orders-api
- paypal-webhooks
- postgresql
- python
- react
- rest-api
- typescript
- webhooks