Descriptions:
Chris Raroque shares a detailed two-month progress report on Amy, his AI-powered calorie tracking app for iOS, with an unusually granular look at AI API economics and the specific engineering changes that improved both unit economics and user retention.
At the two-month mark, Amy reached $1,500 in monthly recurring revenue with 148 paid subscribers at roughly $10 per month, against an AI bill that had dropped from $700 in month one to $221 in month two despite more users. Two changes drove the reduction: routing simple portion edits to Gemini 2.5 Flash Light instead of Perplexity Sonar (cutting per-call cost from half a cent to a tenth of a cent), and adding a Supabase caching layer that returns previously calculated nutrition data for repeated foods. The cache hit rate landed at 60–70% — far above expectations — and accounts for the majority of the savings. PostHog’s LLM analytics feature is shown tracking per-model costs in real time, with Sonar at $217 and Gemini at just $3 for the month.
On the retention side, Raroque tracks week-one retention as his north-star metric — the percentage of users still active seven days after signup. It grew from 3% at launch to 8% after month one, and reached 10% at the two-month mark. A new meal scan feature, using the phone camera to parse full menus and receipts and auto-populate calorie entries, is presented as the most significant bet on pushing that number higher.
📺 Source: Chris Raroque · Published February 09, 2026
🏷️ Format: Workflow Case Study
