Staff Engineer · AI/ML PlatformHyderabad, India
Staff Engineer with ten years of experience taking software and ML systems to production. I have worked on MLOps since 2017, built search and ranking systems along the way, and now build LLM platforms, agents and observability.
10+
years shipping production software
2017
first MLOps pipelines in production
6
domains: cloud, UI, backend, data, ML and LLMs
4
side projects shipped: dev tools and apps
01Focus
01
Model gateways with routing and cost attribution, agent orchestration, RAG, and the evaluation and observability around them.
02
Taking notebooks and model artefacts to inference APIs for classical ML, deep learning and LLMs, on public clouds and air-gapped private GPU clusters.
03
Enterprise search across systems with different auth models and schemas, Elasticsearch relevance tuning, and topic modelling to improve ranking.
02Projects
Kubernetes · Agents · LLM serving · MCP
ActiveKubernetes-native serving for AI agents, self-hosted LLMs and MCP servers. An Agent CRD for A2A, HTTP and framework-backed containers, a Model CRD that turns Hugging Face GGUF weights into an OpenAI-compatible llama.cpp deployment, one gateway, and Prometheus metrics throughout.
Kubernetes · Rust · Tauri · Desktop
v0.1.5An Apache-2.0 desktop client for Kubernetes. It reads your existing kubeconfig and talks straight to the API server, with no in-cluster component and no telemetry. It uses the server’s discovery and table endpoints, so custom resources work without an upgrade.
Mobile · Android · iOS soon
On Google PlayThe traditional Hindu calendar, computed on the device: tithi, nakshatra, yoga, karana and vara for any date and city, with sunrise and sunset, in English, Hindi and Telugu.
Rust · Apache DataFusion · SQL
A SQL engine for querying CSV, Parquet and Delta Lake files, with a desktop GUI and a terminal UI. Built on Apache DataFusion.
03Speaking
July 25, 2026 · Hyderabad
Evaluation and failure detection for production AI systems, where request-level tracing tells you what happened but not whether it was right.
Cloud Native Hyderabad: AI x Observability Meetup
04Writing
Frontier model training has quietly shifted from a chip problem to a power-and-capital problem. A grounded look at the four scarce inputs (compute, power, data, talent) and who should (and shouldn't) build from scratch.
A systems-level deep dive into vLLM: PagedAttention, the scheduler, KV cache management, continuous batching, tensor parallelism, speculative decoding, and the V1 architecture.
In ~30 minutes, Claude Code built a working SQL engine from scratch. This is my reflection on vibe coding with AI — what worked and what didn't.
Learn how to implement distributed locking with Redis to handle concurrent processes in a movie booking system. This tutorial covers building a FastAPI application with Redis for managing shared resources safely.
Keeping track of changes and revisions of your data is more crucial than ever in the fast-paced world of AI, data science, and analysis today. A crucial tool for managing these changes is version control, which makes it simple to keep track of, work together on, and go back to earlier iterations of your data.