Financial Operating System for Household Investing
Co-founding engineer on a consumer fintech product: portfolio tracking, investment research, retirement projection, rebalancing and a natural-language interface backed by a tool-using AI agent that reasons over a household's real financial position. Built as a monorepo of independently deployed apps sharing one design system and one typed database layer, with a confidentiality boundary enforced by the build itself.
Architecture: trust boundary and agent layer
Engineering highlights
- Hard confidentiality boundary enforced at the dependency-graph level rather than by convention, so a privacy mistake fails the build instead of reaching production
- A live tool-using agent that answers questions against a customer's real financial data and shows the evidence behind each answer
- Automated safety tooling in CI, including a pre-commit scanner that blocks confidential figures from ever reaching public-facing code
- Security and compliance groundwork ahead of a regulated open-banking data integration: consent architecture, access-control hardening, runtime monitoring
How it is built
- Engineered almost entirely through AI-native development workflows: every change reviewed by a fresh-context AI pass before merge, shipped through a sequential rebase-per-PR merge queue, and verified against a real running build rather than tests alone
- Two AI development agents now contribute to the same production codebase in parallel, which required explicit ownership and coordination protocols
- Hundreds of pull requests and database migrations shipped to a live system over several months of continuous iteration
Status
Live in production and onboarding its first outside beta customers, with real legal, security and compliance groundwork underway for regulated financial-data access.
Under the hood
Next.js / React monorepo · Postgres · Vercel