Shipping an AI feature to paying customers soon? This is a one-week review that tells you whether it holds up.
I check five things: whether "good" is defined as a number rather than a feeling, whether there is a "never" list for the model, whether evals run on every prompt change, whether there is a cost-per-request budget, and whether you can roll back in minutes.
How it runs: a 60-minute kickoff, then I go through your prompts, evals, failure paths, cost model and rollback plan. You get a written report on what is missing and what to do about it, then a 45-minute walkthrough.
One week, fixed scope, fixed price, a report you keep. Not a code review and not a pentest.
Shipping an AI feature to paying customers soon? This is a one-week review that tells you whether it holds up.
I check five things: whether "good" is defined as a number rather than a feeling, whether there is a "never" list for the model, whether evals run on every prompt change, whether there is a cost-per-request budget, and whether you can roll back in minutes.
How it runs: a 60-minute kickoff, then I go through your prompts, evals, failure paths, cost model and rollback plan. You get a written report on what is missing and what to do about it, then a 45-minute walkthrough.
One week, fixed scope, fixed price, a report you keep. Not a code review and not a pentest.