The founder's story
In 2000 the plan was months of manual rework: six hundred screens of a Medicaid client to be reproduced by hand in a new architecture. I built a code generator instead. Schemas and metadata in, consistent data access, controllers and templates out. It became a reusable company asset and won an award, but the lesson mattered more than the trophy.
The first version was rough. I rebuilt it after watching it run. The second was leaner. That has been the discipline for nearly three decades across enterprise Java, federal modernization, cloud platforms and integration: do not solve the immediate task, solve the class of problem, then refine it on real feedback.
When I started building retrieval systems and agentic workflows, the same failure modes were everywhere. Remarkable models inside weak systems: no architecture, no provenance, no plan for the day the model is slower or wrong. Meanwhile the correctness-critical parts of the business, the calculations, the approvals, the money, were being handed to something probabilistic because it was fashionable.
MetisMesh is the result. Since June 2025 I have split my time between consulting engagements and products I design, build and operate myself. The consulting work spans a treasury knowledge system on Google Cloud, a controlled multimodal review platform in AWS GovCloud, an oncology knowledge assistant on Bedrock, and a provider-search agent integration in healthcare. The products are the six on this site. Both kinds of work keep me close to the whole system: the business problem, the data model, the security boundary, the implementation, the deploy, and what it costs to run.
We do not catch fish for people. We teach them to fish, and we leave behind something that keeps working after we have gone.
Khalid Rizvi, founder