Inspiration

Nigerians don't struggle to earn small money — they struggle to keep it. Existing savings locks are all-or-nothing. We wanted money that stays spendable on its purpose but can't be impulsively cashed out. Hence the name: the Zeroth Law — the AI is the "third body" that keeps you and your money in equilibrium.

What it does

You lock money into purpose wallets and delegate spending to an AI agent you just talk to: "Buy ₦500 airtime", "abeg buy me 1k data", "send ₦5k to Samuel for feeding." It executes real payments — but tell it to "cash out ₦40k" and it's 🔒 blocked, by design.

How we built it

Next.js 14, TypeScript, Tailwind, shadcn/ui on Vercel. Google Gemini (gemini-flash-latest) with strict JSON output does two jobs: parsing natural language into a structured payment intent, and suggesting a personalised money split. A deterministic policy engine — not the AI — makes every money decision.

Challenges we ran into

Telling a peer transfer apart from a cash-out (we ask "what's it for?" to pick the wallet); making the AI split sum to the exact total; and surviving flaky venue wifi + model quotas with an offline fallback.

Accomplishments that we're proud of

A genuinely safe money-agent — the AI never has authority over funds, only language. Multi-turn conversations, Pidgin/slang understanding, an AI-built custom split, and a full monochrome light/dark redesign — all shipped and live in one day.

What we learned

The cleanest way to let an LLM near money is to give it zero authority over money: separate understanding language (AI) from moving money (deterministic rules). That separation is what makes it auditable and demo-proof.

What's next for Zeroth

Real payment rails (VTpass / Reloadly / Flutterwave), persistent multi-user accounts, per-category autonomy budgets, and bank-statement onboarding that infers your split automatically.

Built With

Share this project:

Updates

Submission history