Descriptions:
Chris Raroque gives a candid monthly update on Amy, his AI-powered calorie tracking app built in the style of Apple Notes, sharing real business metrics and the engineering decisions behind two new features aimed at improving week-one retention. Amy currently sits at $2,800 in monthly recurring revenue with roughly 85% profit margins and $400 in monthly costs — almost entirely AI inference. Week-one retention is 13%, down from an earlier miscalculated high, and the episode is organized around the attempts to push it higher.
The main feature is barcode scanning, which turned out to be a two-part problem: extracting a barcode identifier from an image (solved quickly with Apple’s Vision Kit) and mapping that identifier to a product (far more complex). Raroque used Claude Deep Research to survey available APIs before writing any code, ultimately choosing Open Food Facts — a volunteer-curated database of 4 million products across 150 countries — as the primary source. When Open Food Facts returns no results, a fallback lets users photograph the nutrition label directly, with Gemini 2.5 Flash extracting macro data from the image. For labels that wrap around packaging and can’t be captured in one shot, he built a multi-photo stitching flow: Gemini flags when a label looks cut off, prompts for additional photos, and combines them before running the calculation.
The video also covers the gap between testing and real use — bugs in partially torn barcodes only appeared once Raroque used the app with actual groceries — and recommends using WhisperFlow voice dictation with Claude Code to generate more detailed prompts. A practical case study for solo developers building AI-native consumer apps.
📺 Source: Chris Raroque · Published June 01, 2026
🏷️ Format: Workflow Case Study
