• 

  Main dashboard

CloseMate — Finance Operations for Polish Sole Proprietors

About the project

Freelancers and Polish sole proprietors (jednoosobowa działalność gospodarcza, or JDG) often have to reconcile invoices against bank transfers manually. A payment can look correct while still leaving an unexplained difference, an overdue invoice, or a question that needs an accountant's attention.

CloseMate turns that reconciliation work into a clear, reviewable workflow.

The app works with invoice and bank-transaction data across PLN, EUR, and USD. It uses deterministic reconciliation rules to identify discrepancies and potential matches, then uses AI to explain individual results in plain language and suggest safe next steps. Instead of treating an AI-generated explanation as a financial decision, CloseMate keeps uncertain cases open for human review.

What we built

• Multi-currency reconciliation: compare invoice totals with linked payments and surface amount mismatches.
• Human-in-the-loop review: explain why a case needs attention without automatically approving adjustments or initiating payments.
• AI-assisted explanations: use Nicrron to turn reconciliation results into understandable summaries and actionable review suggestions.
• Observability: integrate PRISM tracing to inspect model interactions and investigate failures.
• Accountant-ready exports: generate a CSV with Polish headers, semicolon separators, decimal commas, UTF-8 BOM, and formula-injection protections.
• Polish and English UI: make the workflow accessible to Polish business owners while keeping an English interface available.
• Privacy-conscious design: use synthetic demo data and keep reviewer notes local by default.

How it works

1. Load invoice and bank-transaction records.
2. Match payments and identify discrepancies using deterministic rules.
3. Ask the AI to explain a reconciliation result and suggest safe next steps.
4. Review the explanation, investigate the underlying evidence, and keep uncertain cases open.
5. Export the reviewed data for an accountant.

What we learned

Financial workflows need more than a convincing AI answer. They need traceable evidence, predictable rules, safe failure handling, and a clear boundary between an explanation and an action that affects someone's money.

One of our key design decisions was to keep reconciliation logic deterministic and use AI for explanation rather than letting a language model decide whether a discrepancy is financially acceptable. Observability is also part of the work: tracing needs to help us understand the steps behind a result, not just confirm that a model was called.

Scope and limitations

CloseMate is a hackathon prototype using synthetic data. It does not connect to a live bank account, submit information to KSeF, determine tax deductibility, or replace an accountant. It does not initiate payments or refunds. Its purpose is to help people spot and investigate discrepancies before taking action.

Built for the Money Operations track at MONEY TALKS 2 AI × Finance Hackathon.

Built With

• ai
• automation
• csv
• data-reconciliation
• finance
• fintech
• human-in-the-loop
• internationalization
• llm
• next.js
• nicrron
• observability
• prism
• react
• typescript
• upstash
• vercel