Paycheck Committee
Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

Paycheck Committee is a finance agent for students and young workers who live paycheck to paycheck. Every month it answers two questions. (1) Can I pay every bill on time? A deterministic day-by-day cash simulation finds the lowest balance and any shortfall, then prints the exact fix (e.g. move 9,000 from the emergency fund before Nov 8). (2) How much can I safely invest, and in what? Money goes to bills first, then the emergency fund; only the remainder reaches an investment committee of Muse Spark agents: four researcher agents write evidence-cited signal cards from Taiwan market data, five investor personas (Buffett, Lynch, Bogle, Dalio, Kostolany) score three proposals, a devil's advocate attacks the leader and states what would prove it wrong, and a chair explains the result. The risk officer is a rule engine, not an LLM: it vetoes any plan that spends bill money, sells below cost, adds to flagged high-risk or leveraged positions, breaks concentration caps or uses margin. A live web 'committee room' shows the decision chain as a node graph that lights up as each agent finishes; every node opens the exact Meta Model API request and response, and every run is saved as a trace. Typing one sentence ('dentist bill 9,000 due on the 9th') makes Muse Spark add a structured bill and the whole plan reruns. Everything in the repo (data model, cash engine, risk rules, config-driven committee, Meta API client, web UI, tests, video pipeline) was built during the hackathon; only the concept note in docs/ predates it. Demo data is synthetic and it never places orders.