I took an enterprise RAG platform from working software to an operable service.
Pfizer CoLab · RAG Knowledge Platform
Challenge
Nearly a thousand new staff were arriving ahead of peak vaccine season, and the approved material they needed was spread across separate systems. A retrieval system that answers from anything it can reach is not usable in an enterprise: the service had to constrain retrieval to approved sources and, when an answer was wrong, support an investigation that could establish why. A demonstration does neither of those things.
Action
Delivery planning covered the parts that make a platform a service rather than a prototype: source ingestion under governance, the retrieval workflow itself, evaluation, access controls, the user experience, and the organizational rollout behind it. The operating questions were settled explicitly — which sources are eligible, who may reach what, how an answer gets tested, and who owns an incident when one is raised. The AI Operations Dashboard carried the running state of the service. Stable service-level objectives, assigned incident ownership, and a handover to an operations team were the conditions for calling it delivered.
Result
The platform moved into enterprise use with a service model built for more than 500 users: governed ingestion, a defined retrieval workflow, testing and access controls, and a named operational owner. Approximately 1,000 new staff were onboarded through it ahead of peak vaccine season. It was handed over as a service with support behind it rather than as software that happened to work.
