LatePay 💸 — interactive demo

Every number below is a real TabPFN prediction (open-source tabular foundation model), precomputed on from simulated invoice data. No backend needed — the page only looks up precomputed results.

⚠️ SIMULATED / PRECOMPUTED: TabPFN was trained on a locally generated synthetic invoice history (4,000 invoices, seeded RNG). These are illustrative risk estimates, not financial advice.

1 · Four real client profiles

2 · What-if explorer (precomputed grid)

Move the sliders — each combination reads a real precomputed TabPFN prediction. Other features are held at typical values (repeat client, 30-day terms, email agreement).

Prior late rate: 0.2 Contract strength: verbal/email Invoice amount (USD): 2,500

3 · Batch scoring

Upload a CSV of invoices in the full app (Flask, local TabPFN inference) to score hundreds of invoices at once and download the risk report.

Built Oct 12–18, 2026 · DEV Hacktoberfest AI Challenge, Week 2: Side Hustle · MIT · Model: TabPFN (open-source) · Repo: github.com/arindewangan/latepay