Nava

The problem

Trading interfaces demand structured input — ticker, side, quantity, order type — while people think in sentences. The translation layer is genuinely useful, and it is also the most dangerous thing you can build, because the output spends real money.

The hard problem was never the parsing. It was making an LLM-driven path safe enough that a misread sentence, a stale quote, or a compromised social account can't cost a user money.

Approach

Users describe intent from an iOS client; an LLM pipeline parses it into a typed, structured order that executes against a live brokerage API. Django REST and PostgreSQL on the backend, React Native/Expo on the client — roughly 31,500 lines across nine third-party integrations, taken from an empty repo to App Store submission-ready in two months.

Every safety guarantee is enforced at the execution boundary rather than in the interface, so no transport can route around it.