I'm Peter Yousefi, a software developer specializing in AI operations, prompt architecture, and cloud computing — designing and evaluating robust LLM solutions for real technical problems.
I design and evaluate robust LLM solutions to solve real-world technical problems.
I work across the AI stack — retrieval-augmented generation, prompt architecture, and automated evaluation — paired with the cloud infrastructure that gets it into production. Recently that's meant building tools like an AI campaign generator for small businesses and a civic-notification system that collapses hours of manual coordination into under two minutes.
Before AI operations, I spent time in logistics coordination and society leadership — which is probably why I care as much about the person who has to use the system as I do about the model behind it.
A toolkit shaped by shipping AI systems end-to-end — from model behavior to the infrastructure it runs on.
Systems built to be evaluated, deployed, and trusted — not just demoed once.

Small business owners were losing hours per campaign writing copy, sourcing images, and tying promotions to local events — Lumina turns a business description into a full ready-to-publish campaign in minutes. It directs IBM watsonx (Granite) through a five-step workflow to research nearby events and generate on-brand captions, email copy, and imagery, with every field owner-editable before anything goes live — and a full offline fallback mode so it runs with zero API keys.
View on GitHub →
Land managers were spending 2–4 hours per prescribed burn identifying which municipality a burn boundary touched, tracking down the mayor and fire chief's contact details, and drafting a notification — GeoShield automates the entire chain down to under 90 seconds. It directs an AI model to draft calm, professional civic notifications, with unverified contacts flagged so a human always reviews before sending.
View on GitHub → Built at the IBM AI Builders ChallengeWhere the work happened, and what came out of it.
Open to opportunities in AI operations, LLM engineering, and cloud infrastructure. If you've got a hard problem worth another set of hands on, I'll answer.