Descriptions:
Chris Raroque delivers a candid post-mortem on the worst week his AI calorie tracking app Amy has experienced since launch, covering two infrastructure outages and a security vulnerability disclosure in a video that doubles as a practical guide for solo developers building on AI APIs.
The first outage came from a five-hour Supabase failure that exposed an edge case in Amy’s offline mode: when the backend was fully down, authentication broke entirely, leaving users on a blank white screen with no path to support. Raroque walks through the fixes — a proper degraded-state UI and a status page hosted on entirely separate infrastructure — and explains the mistake of hosting a status page on the same system it’s meant to monitor. Days later, Perplexity Sonar (Amy’s AI provider for calorie calculations) went down for 15 minutes. Raroque describes his manual fallback implementation: Sonar failures automatically route to Gemini 2.5 Flash with Exa as the search provider. He chose manual fallback handling over OpenRouter’s built-in model routing because automatic fallbacks sometimes failed to follow JSON schema instructions reliably.
The video also covers a responsibly disclosed security vulnerability that, had it been exploited, could have caused significant financial damage. Throughout, Raroque references his development stack — Claude Code and Cursor for all coding work, WhisperFlow for voice dictation of detailed prompts — and explains why dictating to AI coding tools produces better results than typing. At filming, Amy had 166 paying subscribers and $1,700 in monthly recurring revenue.
📺 Source: Chris Raroque · Published February 23, 2026
🏷️ Format: Workflow Case Study
