PromptOps
An internal prompt-engineering assistant built on a nine-part framework, benchmarked across Claude, ChatGPT, Perplexity and Gemini so the team gets consistent output whichever model they reach for.
The problem
Every PM on the team was writing prompts from scratch, and the quality swung wildly depending on who wrote it and which model they used. Nobody could tell whether a bad answer was a bad model or a bad ask.
What I built
A nine-part prompting framework — role, context, task, constraints, format, examples, evaluation, escalation, and follow-up — wrapped in an assistant that interviews you for the missing pieces and emits a finished prompt. Same input, benchmarked side by side across four models so we know where each one is strongest.
Outcome
Prompt quality stopped depending on the author. New PMs get usable output on day one, and we now have a shared vocabulary for critiquing a prompt instead of arguing about the answer.