Shivam Kumar

I’m a Software Development Engineer who thrives on turning messy, high-impact problems into stable, measurable outcomes. At Razorpay, I integrated “Viveka” into card alerting systems—automating alert runbook sync and context via GitHub Actions into a Qdrant vector database—cutting manual runbook lookup time by 35% during incident response.

I also strengthen reliability end-to-end: I stabilized the staging devstack for card payments across 7+ gateways and 3+ networks, reducing environment-related test failures by 80%, and built CDC pipelines streaming DB changes into Kafka for ledger and event tracking. I traced and fixed a high-cardinality metrics memory leak (about 800,000 metric combinations), improving memory efficiency by 85%, and I enhanced canary deployments on Spinnaker with progressive traffic shifting and automated rollback triggers—reducing bad-deployment detection time by 40% and production incident risk during releases by 30%. Earlier, as a Prompt Engineer, I trained and evaluated LLMs across algorithmic coding and mathematics tasks, improving the quality of 1000+ AI-generated responses across iterations, and contributed to domain-specific fine-tuning. I enjoy working at the intersection of backend engineering, observability, and AI tooling—shipping pragmatic solutions that make teams faster and systems safer.