Skip to content
Wunderlich Web

01. FINANCIAL SAAS · 2024–PRESENT

Nova Net Worth

An AI-first net worth tracker for modern households. Built from scratch: Next.js 15, Supabase, Stripe, Plaid. A 15-layer AI safety pipeline protects every AI-assisted feature.

Nova Net Worth homepage — AI-first net worth tracker

TIMEFRAME

2024 – present

ROLE

Founder, sole engineer. Design, architecture, every line of code.

Nova Net Worth — a personal finance SaaS built for households who want honest, AI-native net worth tracking without the clutter of legacy tools like Mint or Personal Capital.

THE PROBLEM

Every existing net worth tracker either bolted AI on as a chat feature (Personal Capital, Monarch) or ignored it entirely (Mint before shutdown, YNAB). None treated AI as a first-class assistant with guardrails strong enough to talk about real money. The opportunity was an AI-first tracker where the AI is trustworthy enough for a family to actually rely on.

THE APPROACH

  • Built on Next.js 15 + Supabase + Stripe + Plaid from day one — no template, no low-code shortcuts.
  • Designed a 15-layer AI safety pipeline (kill-switch → prompt-injection guard → DLP → PII redaction → pre-check scorer → intent parser → compliance rules → semantic risk scorer → hallucination guard → LLM-as-judge → human review sampling → audit log → facts grounding → budget controls → usage tracking). Each layer is tested independently and documented in its own spec.
  • Wrote a deep test suite from week one — unit, integration, E2E (Playwright), and a 100-case Charlie evaluation suite that scores AI safety on every PR.
  • Ship discipline: 1,600+ PRs merged, typecheck + lint + test + build required on every commit, first-class failover across Claude / GPT / Gemini providers.

STACK

  • Next.js 15 (App Router, React 19)
  • TypeScript (strict)
  • Supabase (Postgres + Auth + Storage)
  • Stripe (billing + dead-letter alerts)
  • Plaid (bank aggregation)
  • Playwright (E2E)
  • Vercel (hosting)
  • Sentry (observability)
  • Claude / GPT / Gemini (multi-provider failover)

RESULTS

  • 15

    AI SAFETY LAYERS

  • 6,000+

    TEST SUITE

  • 84 / 100

    CHARLIE EVAL BASELINE

  • production-grade

    STACK