AI product engineering
Different from bolting AI onto something that exists. Here the AI is the product, and someone has to own the whole thing: what the model should and shouldn't be allowed to decide, the pipeline behind it, the interface around it, the billing, and the machine it runs on.
I've shipped three of these end to end. Auto Screener Agent scores candidates against a role's real criteria, ranks a shortlist and drafts rejection emails for everyone else. Zynkbot AI is a no-code RAG platform — document ingestion, vector search, a management dashboard, subscription billing, and a one-line embed script for any site. Portfolio Roast gives deliberately blunt critique and returns first byte in under half a second, because a slow roast isn't funny.
What this includes
- Product definition — scoping what the model decides, and what it never does
- The full pipeline: ingestion, retrieval, prompting, evaluation
- Interface, dashboard and onboarding
- Billing and multi-tenancy when it's a real product and not a demo
- Deployment, and running it once it's live
Typical stack
Where I've done this
"Hire Smarter, Not Harder" — an AI resume-screening agent that reads, scores and shortlists candidates, and drafts rejection emails for the rest. Built and deployed solo.
"AI-Powered Chatbots for Business Growth" — a no-code platform to spin up chatbots trained on your own knowledge base, with a universal embed script for any site.
"Face the Heat" — an AI portfolio and resume roast that gives blunt, specific, senior-dev-style feedback instead of polite platitudes. Shipped solo.