Platform & Infrastructure at Evinced
Took 2 teams and 11 people to 7 teams and 40+ across five countries, stood up GenAI, ML, Data Infra, and DevOps from zero, and cleared FedRAMP readiness in six months to unblock a $10M government pipeline.
- Platform
- Infra
- GCP
- Kubernetes
- GenAI
- FedRAMP
I own the engineering platform underneath a Series C accessibility company: Core Platform, Data Infra, GenAI, Machine Learning, DevOps, and IT. Forty-plus people, seven engineering and data-science managers, five countries.
The problem
When I took over, R&D was a set of siloed groups handing work across walls. Researchers sat apart from the engineers who shipped their ideas, ownership was blurry, and the platform could not keep up with the validation traffic the product generated. The "Cloud A11y Validations" tier was throttling under its own success, with rate-limiting bottlenecks capping how much the product could scan. The instinct in that situation is to hire. I do not start there.
What I did
I changed how the org worked before I changed its size.
- Restructured silos into autonomous squads with clear ownership, and embedded researchers directly inside engineering teams so experimentation and production ran on parallel tracks instead of in sequence. That shortened the path from research spike to shipped feature and cut cross-team dependencies.
- Built GenAI, ML, Data Infra, and DevOps from zero. The GenAI/ML team now ships intelligent accessibility detection and remediation, and internal MCP-integrated LLM agents handle CI/CD triage, documentation, and first-pass code review so senior engineers stay on product.
- Re-architected the validation tier to handle millions of concurrent requests. I traced the throttling to a shared connection pool and a hot rate-limiter, then sharded both. The cap that bounded product growth is gone.
- Modernized the infrastructure: migrated to GCP Autopilot (GKE) with GitOps delivery on Argo CD and Terraform, and stood up SLA/SLO and observability frameworks across the org.
- Drove FedRAMP readiness in six months, owning the security and compliance posture end to end.
The outcome
- Scaled from 2 teams / 11 people to 7 teams / 40+ across five countries by fixing ownership, not by throwing bodies at the problem.
- The validation tier now absorbs millions of concurrent requests with the rate-limiting bottleneck cleared.
- FedRAMP readiness unblocked a $10M US-government sales pipeline.
- A GCP migration plus environment standardization and vendor consolidation cut ~$120k in annual spend with no hit to delivery velocity.