From protocols
to models.

I'm a software architect with a background in distributed systems and networking, increasingly focused on where AI meets production infrastructure.

The UC Berkeley MIDS program was a deliberate pivot. After two decades in systems engineering, I wanted to build a rigorous foundation in machine learning — not just use the tools, but understand them. The coursework spanned statistics, machine learning, data engineering, and research methods. The capstone project — deciphering ancient Japanese script using OCR, unsupervised learning, and GPT-4 — was representative of what drew me to applied AI: constrained problems that require combining multiple techniques to get anywhere.

At Applied Intuition I worked on the Data Explorer product, which helps autonomous vehicle teams visualize real drive logs and triage them — identifying events of interest and exporting curated collections for further analysis. I built parallel, auto-scaled export workflows and integrated CLIP-based multimodal models to enable natural language search over real-world drive data. It was my first sustained exposure to multimodal ML in a production setting.

My current focus is observability and reliability for complex systems, using agentic AI to reason over telemetry, surface anomalies, and assist with incident response. The hard problem isn't building the agent — it's knowing when to trust it, when to escalate, and how to keep a human meaningfully in the loop.

After nearly 25 years in the US, I returned to Bengaluru in 2024. The problems that interest me most sit at the boundary of AI, software systems, and the physical world — where the stakes are real and the margin for error is small.