Cortex — AI finance tracker with natural language input
AI finance tracker with natural language expense entry, custom Pomodoro timers, and a botanical interface.

Live demo
The running product, embedded below. If it refuses to load in an embed, the live site opens in a new tab.
Problem
Expense apps die at data entry — people stop logging the moment the form feels like work.
Solution
Cortex parses natural-language spending through the Gemini API and Groq, persists the ledger in Supabase PostgreSQL, and pairs it with custom Pomodoro focus timers in a botanical Next.js 15 interface.
Decisions I made
- 01Logging is a sentence, not a form — expense apps die at data entry; 'spent 140 on lunch with team' parsing into a ledger row is the entire product bet.
- 02Two LLMs, one job — Gemini does the parsing, Groq is the fast-path; routing per call keeps latency low without a bespoke model.
- 03Postgres as the ledger — money data belongs in real, queryable, backed-up tables, not in vector embeddings or app state.
- 04Focus layer in-app — a custom Pomodoro that runs alongside the ledger keeps daily logging attached to a routine instead of a chore.
Outcome
A production finance tracker where logging is a sentence, not a form — plus a focus layer that keeps the app part of the daily routine.
This project is part of the work I ship as a full-stack developer — see the freelance services page for project types, process, and how to start.