AI reliability audits and embedded engineering for software teams
Software companies, SaaS products, AI startups and technical agencies.
You shipped an AI feature and have no way to answer "is it right?" beyond spot-checking it. Nobody wants to be the one who finds out from a customer.
What I can actually show you
A public multi-tenant platform reference with the isolation tests that enforce it, and a public unauthenticated endpoint with 14 SSRF vectors blocked and tested.
What I build for technology & ai products
An evaluation harness for what you already shipped
Your prompts, your retrieval and your failure modes reviewed, with an eval-harness design and a prioritised fix list. Fixed price, about a week, no retainer after it.
The reliability layer underneath the feature
Retrieval gates and checks that catch a wrong answer before a user sees it — the layer I built for a HIPAA platform where a wrong answer reaching a patient was not survivable.
A forward-deployed engineer for your messiest customer
I embed with your team, learn the customer's domain, and ship the integration that makes your product work in their world.
Where this usually starts
AI Reliability Audit — $2,500
Published pricing, fixed scope, and a free Operations Audit first so we both know whether the build is worth doing before either of us commits to it.
Questions I get asked
What does the AI Reliability Audit actually deliver?
A review of your prompts, retrieval and failure modes, an eval-harness design you can implement, and a prioritised fix list — in about a week for a fixed $2,500. No retainer and no commitment past it.
We have engineers. Why bring in someone else?
Usually because they are busy shipping features and evaluation is the thing that never gets prioritised until something goes wrong. An outside week spent solely on "how would we know if this were wrong" is cheap next to finding out from a customer.
Do you take embedded or forward-deployed contracts?
Yes — that is the work I most want. There is a page about exactly how I approach it at /engineering, including the public code and three production bugs worth reading.