How We Built a Weather-Aware Daily Recipe Pipeline With FastAPI

Fancy Fox publishes one AI-generated seasonal recipe idea daily. The pipeline consists of distinct steps: gathering context, generating a recipe, validating the output, rendering static pages, and promoting the result. The key engineering challenge was determining which steps should fail, degrade gracefully, or be recreated on demand.

The pipeline is structured as a state machine, ensuring that a malformed recipe stops publication, while optional promotion does not erase a valid recipe, and browser failures report issues instead of claiming success. The model output is converted into a Pydantic model before being used in templates or storage, normalizing strings and handling empty values.

Optional work returns honestly by returning silently when no eligible item exists and logging promotions only when the browser action is successful. Browser state is treated as disposable, with a lifecycle that connects to a controlled browser or launches a fresh instance. Operational logs are captured using a rotating file handler, ensuring that logs persist beyond terminal scrollback.

Static output benefits from a FastAPI development and orchestration surface, with the public website rendered as static HTML for stable URLs, complete metadata, and crawlable links. The service currently features over 1,500 pages, with a curated sitemap of 50 URLs. Future improvements involve connecting recipe impressions to save, return visit, or app open, then using these signals to enhance the featured set.