18+ years building DevOps, SRE, cloud and secure delivery platforms across financial services, media and enterprise tech. Now applying GenAI, RAG and developer tooling to engineering problems at scale.
whoamiakshay-sinha · platform eng, devops & ai · 18+ yrsrag ingest --from jira,incidents,docs✓ embeddings stored in postgres · assistant onlinekubectl apply -f gatekeeper/constraints/✓ opa admission policies enforcedls ~/stackgenai/ kubernetes/ opa/ terraform/ airflow/
RAG assistants, defect intelligence and agentic developer tooling built on foundation models.
Kubernetes and OpenShift platforms with policy-driven admission control and automated provisioning.
Infrastructure and policy as code, with security controls that are validated automatically.
Pipelines, orchestration and APIs behind critical, high-volume trading workflows.
Leading DevOps, automation and production engineering for Equity Swaps platforms in Institutional Securities.
Led DevOps teams for PepsiCo and Universal Studios. Built a global NGINX API platform and ran CDN operations on Akamai and Cloudflare.
Led DevOps for Comcast, NBC Universal and Universal Studios. Automated API publishing, microservice onboarding and large-scale data migration.
Automated build, release and deployment for WebSphere, Java and .NET applications.
Lowering one inherited learning rate produced a stable continual-learning baseline that passed every gate, but extra replay storage bought nothing measurable and the last precommitted diagnostic returned NO_RATE.
A continual-learning grid finished every cell and reported the calibrated configuration while running a different update dose; the repaired rerun still failed cross-seed stability in 18 of 24 groups.
Frozen word embeddings failed a relation-extraction test, a supervised probe blamed the readout, and an order-aware probe then showed the probe had most likely been reading topic, not relations.
Whether it's a delivery platform that needs rebuilding, security controls that should run themselves, or an AI assistant grounded in your own engineering data, I'd love to hear about it.