← Back to the hackathonThe track
Agents in Finance
Build a team of AI agents that does a real job in finance or fintech. Pick a problem where money, risk or paperwork is involved, and show several agents working on it together.
Where to look
These are suggestions, not a list to choose from. Bring your own problem if you have one.
- Payments and checkout. Agents that watch a store's Black Friday payments as they come in: flag suspicious orders, answer refund and chargeback requests, keep the ledger right.
- Back office. Match invoices to payments, chase what is missing, close the books.
- Risk and compliance. Review onboarding documents or transactions, with one agent checking another's findings against the source.
- Personal finance. Agents that read a person's statements and help them budget, save or plan.
- Research. A team of analysts: one reads filings, one follows the market, one writes the note, one checks every number in it.
Ground rules
- Synthetic or public data only. No real customer data, no real accounts, and no real money moving.
- Every number has a source. A figure an agent reports points to where it came from. Invented numbers cost points.
- Not financial advice. Where a user could take the output for advice, the system says that it is not.
How it's scored
| Criterion | Points |
|---|
| Useful: it solves a finance problem someone has | 30 |
| Multi-agent design on Mozaik: agents in parallel, shared state, checks between them | 30 |
| It works: we can run it from the README | 25 |
| Care with financial data: sources shown, no invented numbers | 15 |
For every team
- Built on Mozaik. Every team. One agent with a pile of tools doesn't count.
- Deadline: Friday 27 November, midnight CET. Resubmit as often as you like; the last entry counts.
- Submit: a public GitHub repo with a README: what it does, the architecture, how to run it. No video needed.
- Prizes: $500, $300 and $200 for the top three. Each entry is scored out of 100 on the criteria above.