Burrfect AI Pipeline

> I built a working AI prototype that pulls roaster and bean data from the web, structures it for our database, and publishes espresso recipes to our website -- creating a growth loop between our app and the open web. ## The Problem Burrfect's recommendation engine needs rich data about coffee roasters and beans -- tasting notes, origins, roast profiles -- but manually researching and entering that data doesn't scale. We needed a way to automatically source, structure, and publish this information, both to power the app's recommendations and to attract new users through search-discoverable content on our website. ## My Role I prototyped the AI pipeline end-to-end, wrote the PRD for production implementation, built the iPaaS automations that connect the pipeline to our WordPress site, and configured the site to display the data -- while my developer Aayush translated the prototype into production Cloud Functions. ## The Approach I started where I always start: with a working prototype I could iterate on fast. Google Apps Script and Sheets gave me a zero-infrastructure environment to chain Perplexity API calls (for web research) into OpenAI calls (for structured extraction) and validate the output against our schema before touching the database. The goal wasn't production code -- it was proving the data quality was good enough to trust. Once the pipeline logic was validated, I wrote a detailed PRD with my working prototype as a reference implementation and handed it to Aayush, who rebuilt it properly in Google Cloud Functions with Firestore as the backend. This is the pattern I keep coming back to: I prototype fast, validate the concept, then spec it clearly enough that an engineer can build the real thing without ambiguity. The piece I kept ownership of was the growth loop. I built Albato automations using webhooks so that every time a new roaster, bean, or improved recipe surfaces in the app, it's immediately published to the Burrfect website. I configured the WordPress site to accept and display all of that structured data. The result is a flywheel: app users generate better recipes, recipes publish to the website, the website attracts new users through search, and those users download the app. ## What I Built - **AI data pipeline prototype** -- Perplexity API for web research, OpenAI for schema-specific JSON extraction, validated against our database schema in Google Apps Script - **Production PRD** -- Detailed spec with working prototype code, enabling clean handoff to developer for Cloud Functions implementation - **iPaaS automation layer** -- Albato workflows with webhooks that push new roasters, beans, and recipes from the app to WordPress in real time - **WordPress data display** -- Configured the Burrfect website to accept, format, and publish structured espresso recipe and roaster data - **Content growth loop** -- Automated pipeline from app-generated data to SEO-discoverable web content ## The Result Live in production. Every roaster and bean that enters the Burrfect ecosystem is automatically enriched with AI-sourced data and published to the website. Every time a user dials in a better recipe, it updates the espresso recipe book on burrfect.io. The site now functions as both a marketing channel and a reference resource, with content that grows organically from app usage -- no editorial team required. This project demonstrates what I think the AI Product Builder role actually looks like: not waiting for an engineering sprint to explore whether something works, but building a functional prototype, proving the value, writing a clear spec, and then orchestrating the pieces -- APIs, automations, integrations -- that turn it into a production system. ## Tech Stack - **Prototype:** Google Apps Script, Google Sheets - **AI APIs:** Perplexity API (web research), OpenAI API (structured extraction) - **Production backend:** Google Cloud Functions, Firestore (built by developer from my PRD) - **Automation:** Albato iPaaS, webhooks - **Publishing:** WordPress (burrfect.io)