Building
Systems
that run in production

AI Software Development Engineer, Software Engineer – AI Systems, and MLOps / Platform Engineering roles where I own the agentic orchestration layer and the production infrastructure beneath it. I'm happiest building multi-agent systems, retrieval and natural-language query engines, and the automation that ships them reliably.

Yes. I'm based in Bangalore, India and open to hybrid, remote, and relocation for the right team, including global opportunities working across time zones.

I architect for reliability, not demos: 7 production AI systems and 8 Kubernetes-native microservices at HPE, 16 legacy backend services migrated to cloud-native architectures with standardized deployment and rollback, 750+ CI/CD deployments across 12+ microservices serving 500+ organizations, 60% API performance gains through query and indexing work, and MLOps pipelines that handle deployment, monitoring, and lifecycle management end-to-end.

The whole pipeline: multi-agent orchestration (LangGraph, LangChain, AutoGen), retrieval and embeddings, prompt engineering and RLHF feedback loops, plus the cloud-native infrastructure (AWS, Azure, Docker, Kubernetes) to run it in production.

By treating autonomy as an engineering property, not a prompt trick. I build agents on LangGraph with typed tools and deterministic skills, fan-out routing for parallel work, and explicit guards around every model call: timeouts, retries, and schema validation. Reasoning steps are logged and evaluable, so failures isolate and debug cleanly instead of turning mysterious. That's how RapidDoc's multi-agent pipeline indexes large monorepos, and how Hexus pins root cause across noisy multi-tenant logs.

End-to-end ownership: containerized model serving on Kubernetes / OpenShift, automated deployment and rollback, monitoring for drift and latency, and RLHF feedback loops that fold real usage back into evaluation. At HPE I build the pipelines that handle deployment, monitoring, and lifecycle management across hybrid-cloud environments, plus the sustainable automation around them, which improved infrastructure visibility and removed 14 engineering hours of manual analysis every week. A model that improves in a notebook actually reaches users safely.

I scaled a B2B health-tech ERP to 500+ organizations, so tenant isolation and performance were daily constraints. That meant careful data partitioning, indexing and query tuning that cut API latency 60%, and 750+ CI/CD deployments kept safe with automated tests and 50% fewer regressions. I design for failure modes first: isolation, observability, graceful degradation. Then optimize.

Yes. I was Principal Author on a published blockchain-based recommendation system (ICT4SD 2025, Springer) and have placed in multiple national competitions: winner of HPE's FOCUS AI Ideathon 2026, runner-up at HPE Digithon, top 8 of 200+ teams at IEEE Kalpana, top 4 in a digital-forensics CTF and #60 globally of 500+ teams in Nexus CTF.

I'm available to talk now. The fastest way to reach me is the contact form below or email directly. I usually respond within 24 hours. My full experience, projects and skills are on this site, and the résumé page has the current status.