Kombocode is an AI/ML engineering practice for payer-tech, care-management, and clinical-trial systems. We stabilize what's breaking in production and architect what's next.
A claims or prior-auth model is live. Decisions go out every day. Then CMS asks why the model denied a specific case last quarter — and the team can't reconstruct it. The model isn't broken. The audit trail is.
An ML pipeline ships predictions in production. The AWS bill is 3× what was budgeted. Half the spend is retry traffic from a feature that quietly degrades every Tuesday. Engineering is told to optimize without taking the system offline.
Most healthcare AI fails audit not because of the model. It fails because nobody can reconstruct the decision. That's an engineering problem.
I built this to answer one question: when a payer AI denial decision is challenged — by a regulator, an audit, an appeal — can the decision actually be reconstructed?
The diagnostic runs on synthetic scenarios drawn from real payer workflows — prior auth, claims AI, UR nurse review, appeals replay — and walks through whether the denial path can be reconstructed end-to-end within ten minutes. It exposes the trace gaps most production systems quietly fail.
It's best seen walked through on a call.
Engineering, not advisory. Hands on the system, not slides about the system.
Reconstruct what the model actually did, at decision granularity. If a single decision can't be traced under audit pressure, nothing downstream matters.
Find where production diverges from the lab: silent feature drift, retry storms, runaway cost, edge cases that never made it into the test set. Close those gaps before they become incidents.
Get the system to the state where a CMS request or an internal compliance review is a 24-hour answer, not a six-week scramble.
Two engagements, anonymized by sector and stage. The transferable pattern: rescue and stabilize regulated-healthcare AI in production.
I founded Kombocode after twenty years building software for production in regulated environments, the kind where a system failing is a serious problem, not just a quick fix. I've spent most of that time in Python, Java, Kotlin, React, Angular, and AWS. These days my focus is AI and ML: the systems healthcare platforms are starting to put real decisions behind.
Every engagement runs through me. I lead the discovery, own the architecture, and sign off on every technical review. The build happens under my direction, and the result is mine to stand behind.
Thirty minutes, on the phone. We talk about what's live, what's fragile, what's about to be audited. If there's a fit, we go from there.
