Record
Selected work, anonymized. Each note says what the situation was, what changed in the system, and what moved. The figures are the ones I would stand behind on a reference call.
Turning alert noise back into signal
Situation. Production sent the team 104 alerts a day. Most were false alarms or flaky, intermittent issues that needed no action. Each one still meant documentation, a root-cause write-up or a meeting to justify it, and the alerts that mattered sometimes got missed.
What changed. My team and I redesigned the alerting logic so an alert fires only when it needs a person. The team got its time back, and real issues stopped hiding in the noise.
Claims triage that survived its pilot
Situation. Claims adjudication was manual and siloed. People checked every claim's data for completeness and correctness, then adjudicated it by hand.
What changed. My team and I automated the pre-adjudication checks and brought in machine learning models to adjudicate. The models now make the first call and the team reviews their work, keeping a human in the loop.
Underwriters reviewing, not typing
Situation. Underwriting was manual. Underwriters spent much of their time typing long documents.
What changed. My team and I brought in a GenAI assistant on a secure retrieval architecture. Underwriters now oversee and review its work instead of writing from scratch.
Cost as an execution outcome
Situation. A set of nearly identical applications served users around the world. Each geography still ran on its own dedicated infrastructure.
What changed. My team and I re-architected the applications and infrastructure to share resources across geographies, with no loss of performance. Reworking logging and alerting cut the infrastructure needed further. The applications are now easier to maintain.
Partner onboarding as an experience, not a deal
Situation. Partner onboarding had no standard procedure. SLAs existed on paper but were not honored. Onboarding was slow and uncertain, and some partners went quiet or dropped out.
What changed. I built the system around it: SOPs, a known-error database, guides for partners and new team members, reusable templates, integration documents and FAQs. I scripted the database entries, set a check-in cadence with partners and aligned stakeholders on one approval process. Queries got resolved far faster, and every partner followed the same path.
From hours billed to product delivered
Situation. The engagement ran as staff augmentation, measured in hours burned. A lot of the team's time went into estimates: writing them, defending them, documenting them and renegotiating them, on top of the work itself.
What changed. I introduced the Product-Oriented Delivery model to the practice. The engagement was then measured on the product delivered, not the hours or headcount behind it. The team stopped carrying the estimation load and put its attention back on the work.