Life Finance

Make the mistake here first. A 3D financial-learning game where AI brings the lesson to life—but never controls the money.

Inspiration

People often learn finance only after a repair bill, layoff, or credit-card balance makes the lesson expensive. Life Finance lets learners feel those tradeoffs before real money is at risk. It turns advice into experience, so a bad choice becomes feedback—not a failing grade.

What it does

Life Finance is a 3D personal-finance strategy game played one month at a time. Players choose a persona, plan, face life events, and watch each choice change cash, debt, taxes, investments, and net worth. Sprout guides them while Debtzilla, Inflato, and Impulso personify financial pressures.

Our promise: AI can improvise the story, but it cannot hallucinate your money. Local ML ranks only approved events; optional language features interpret or narrate. A deterministic engine owns every dollar, with button fallbacks for uncertain input.

How we built it

Next.js, React, TypeScript, and Three.js power the experience. PostgreSQL, Drizzle, and Supabase support saves; an instant demo needs no database or API key.

An exact-cent engine owns taxes, debt, investments, events, and outcomes. A self-trained local ranker covers 648 queries and 3,240 candidates, runs without an API call, and can never bypass engine safety checks.

Codex with GPT-5.6 was a core building partner. It accelerated the architecture, finance engine, ML pipeline, adversarial tests, simulations, debugging, and Sprout’s UI. We kept the key decisions human: AI never owns the math, every dollar stays reproducible, and gameplay teaches through consequences.

Challenges we ran into

One confident but incorrect number could undermine the lesson, so we separated AI personality from deterministic financial authority. We also had to make difficulty educational rather than arbitrary. Replayable randomness, safety checks, and a Balance Lab tested whether prepared and reckless strategies diverged for understandable reasons. The last challenge was hiding that complexity beneath clear choices and friendly characters.

Accomplishments that we're proud of

• A runnable product: judges can plan, advance months, resolve events, and watch the board update.
• Enforceable AI limits: no model can invent a charge or rewrite a balance.
• Local, constrained ML: ranking works offline with deterministic fallback.
• Testing as design: Codex helped build invariants, adversarial tests, and long simulations.

What we learned

The most useful AI role is not always the most powerful: AI interprets and explains; deterministic software calculates and enforces. We also learned that education needs recovery. Players build intuition when they can understand a mistake, prepare differently, and make the same challenge manageable.

What's next for Life Finance

Next is Nemesis Mode: an opt-in rival that studies evidenced weaknesses and selects a useful engine-approved challenge. Debtzilla might target a thin emergency fund, explain why, and lose power as the player improves—while the engine still locks every number.

We also plan to add events, debriefs, counterfactuals, personalized goals, and classroom modes for comparing strategies without exposing real finances.

Built With

• ai
• chatgpt
• codex