I'm an engineer and ML team lead based in San Francisco. At Rubrik, my team builds SAGE, the Semantic AI Governance Engine for Rubrik Agent Cloud. These days, I'm focused on agentic security: how we govern what agents do as they do it, and make them safe enough to trust with useful work.
I grew up in India, where a good education could change someone’s life, but the chance to get one wasn’t available to everyone. Cost mattered. So did access to good teachers, and whether you could go to school at all.
That’s a big part of why I’m excited about AI. I think it can bring learning to people who haven’t had those opportunities, perhaps even more widely than the internet did. But that only works if people can afford to use it. My work on making AI cheaper and more efficient to run is one way I can help.
I also want what we learn to be useful beyond our own team. So I work in the open when I can, publish our work, and share how we’ve built these systems at conferences.
I co-created Turbo LoRA, which combines fine-tuning with speculative decoding. We wanted faster inference without giving up task accuracy.
Open source · PythonI’m a core maintainer of Ludwig, an open-source framework that lets you train models by describing what you need in a configuration file.
Research · 2024We tested 310 fine-tuned LLMs across 10 models and 31 tasks. I co-authored the report on what we found.
Travis Addair and I teach this short course on reinforcement fine-tuning. We walk through GRPO and how to design rewards for the behavior you want from a model.
Take the courseI explain how Turbo LoRA works and how we use it to speed up inference for fine-tuned models.
Arize:Observe · 38 min · 2024My talk at Arize:Observe on fine-tuning and serving smaller models with Ludwig and LoRAX.
I write about the systems we build and what I’m learning at work and outside it.
I spent three years at Predibase. Here’s what that taught me about choosing a startup to join.
We need to check what agents do without slowing them down. This piece covers how we approach latency and accuracy in SAGE.
We used compilation and correctness as rewards to teach a model to turn PyTorch code into Triton kernels.
I’ve been thinking about how AI changes the work of an engineering manager, and where context becomes a bottleneck.
Tech Lead Manager, AI
Previously Senior ML Engineer · Promoted Oct 2025
I lead the team building SAGE for Rubrik Agent Cloud. We run small language models at high volume to enforce policies as agents work. The challenge is keeping latency and cost low while maintaining high recall.
ML Tech Lead
Promoted Jun 2024
Previously Senior ML Engineer · Promoted Sep 2023
Joined as ML Engineer · Jun 2022
Much of my work was about making smaller models useful and cheaper to run. I worked on training and inference, helped maintain Ludwig, and co-created Turbo LoRA. I also co-authored the LoRA Land report. We were acquired by Rubrik in 2025.
ML Scientist, Core Machine Learning
Previously Associate ML Scientist · Promoted Apr 2022
My first role after college. I worked on ML features for Confluence and Trello, including building a graph of user interactions to recommend people and spaces to follow in Confluence’s redesigned social feed.
B.S. in Computer Science · Co-founder, DataRes
I co-founded DataRes because I wanted more people to share my excitement about ML with. People I could learn and build with.
I love live music. Jacob Collier was especially memorable because he got the crowd singing like an orchestra. I’ve seen Young the Giant three times, really enjoyed Cage the Elephant, and love Polyphia.
I play guitar too. Mostly classic rock and Metallica, though lately I’ve been learning Hozier and Death Cab for Cutie.
I’ve been to over 25 countries. Italy is a favorite. The Uffizi, Cinque Terre, Rome, Bologna. But when I think about the trip, I also think about how friendly everyone was and the friends I made along the way.
Lately I’ve been making a lot of lattes on my Breville. Turns out there’s quite a bit to get right. I’m still learning how the beans, measurements, and equipment affect the result.
As a kid, I wanted to be an astronaut. Space still gets me. I find it hard to imagine just how big it is, let alone how much we haven’t discovered yet. It makes me curious in a way nothing else does.
Community means a lot to me. I love bringing people together, whether that’s hosting, going to a show, or building a group around something we’re interested in. Mostly, I want it to be a place where people can be themselves.