Team Venorse

PayRecon — Payment Reconciliation Agent Freelancers and small studios who invoice overseas clients get paid in foreign currency, but what lands in their INR account almost never matches the invoice. Intermediary banks skim fees, the FX conversion carries a spread, and bank charges plus GST come off the top. The result is a pile of "close but not equal" payments and no clear answer to the only question that matters: did I actually get underpaid, or is this just fees? PayRecon ingests invoices and bank/platform statements, automatically matches payments to invoices, and explains exactly where every penny went — intermediary fees, FX spread, bank charges, GST — then drafts the follow-up email for whatever is genuinely still owed. It runs one tight end-to-end loop: upload → auto-match → explain the gap → draft follow-up. The core design principle is money math is done in code, never by the model. Every figure on screen comes from pure, unit-tested TypeScript (tested across 1,000 generated cases). The model does the three things models are genuinely good at — reading messy documents, resolving the hard matches as a tool-using agent, and writing plain-language explanations. A number validator then rejects any model-written sentence that introduces a figure not already in the computed breakdown. That's the credibility story for a financial tool: the AI can explain the numbers but can never invent them. Deterministic-first by design: CSV parsing, rule-based matching, and gap math make no model calls and are instant. The agent only runs on the genuinely hard cases — split transfers, combined invoices, instalments, parent-company remitters — streamed live over SSE so you watch each tool call resolve. Built on Next.js (App Router) + Tailwind, Neon Postgres via Drizzle, Better Auth with a one-click "Try the demo" account, Vercel Blob for files, and ECB reference FX rates. The whole app is clickable end-to-end at $0 in stub mode before a single live call is made.