.png&w=256&q=75)
🤓 Latest Submissions
Prism
Small businesses rarely fail because revenue disappears — they fail because owners make cash decisions without seeing the downstream consequences. Project Prism replaces the static dashboard with an AI boardroom: when a financial risk is detected (e.g. a looming payroll shortfall), two AI executives — a CFO focused on liquidity and a Collections Manager focused on receivables — reason sequentially over the business's real numbers and argue opposing strategies. Critically, the AI never invents a number. Every figure — cash balance, runway, payroll gap, projected outcome of each decision — comes from a deterministic TypeScript engine that is fully unit-testable and unchanged by the AI layer. The two executives (powered by Fireworks AI) only generate the natural-language reasoning and counter-argument on top of that fixed financial state, with the second agent reading and rebutting the first's literal output. If the AI is unavailable, the app falls back to an equivalent static boardroom with identical numbers — the demo cannot break. Beyond the live app, an AMD Instinct GPU (ROCm + PyTorch) notebook runs a 50,000-path Monte Carlo scenario analysis per decision option, adding a risk distribution around the deterministic expected outcome, plus a small synthetic-data classifier that validates the risk ranking. The owner stays in control throughout: they see the debate, pick a strategy, and watch the simulation update cash, runway, and risk level instantly — turning financial planning from retrospective reporting into rehearsed decision-making.
13 Jul 2026
