About Me

I am a research scientist at 10x Genomics in California. My research journey includes a postdoctoral fellowship at the Microsoft Research New England Lab with Lester Mackey, a PhD from Stanford’s Information Systems Lab, where I was advised by Professor David Tse, and a masters in Electrical Communication Engineering from Indian Institute of science with Professor P Vijay Kumar. My work has consistently had the theme of designing and deploying scalable machine learning algorithms to solve real problems, and showing their goodness, with problems often coming from computational genomics.

Current Work

At 10x Genomics, I work on developing algorithms and machine learning to efficiently process large amounts of data generated from single-cell-RNA-seq and spatial-transcriptomics in a principled manner. Single-cell-RNA-seq and spatial-transcriptomics present statistical challenges because experiment design is in the setting is hard; and the data generated is mostly observational. In addition, the scale of the data presents algorithmic challenges.

I work in a with people with diverse backgrounds in a team developing the products. In particular I worked on the Visium HD product, where we detect genes on different spatial areas of a tissue slice. I was involved with most aspects of the product development; from interfacing with other teams to model what we were trying to build, to figuring out metrics we had to optimise for a good product to developing algorithms to process our data. The data we generate can be thought of as an image with around 20,000 channels – each corresponding to a gene; and each being sparse in addition to a H&E image. Processing this presents us with many interesting algorithmic and statistical problems.

I have also worked in and led teams that developed and shipped machine learning models from end to end – most notably in the context of annotating cell-types in singe cell data; and in annotating nuclei from microscope images stained with H&E. In both those cases, we worked on end-to-end development of the product; from figuring out what data we would need; to setting up data collection, cleaning and labelling; to building, packaging and shipping the model.

Working in the industry got me writing code and thus in programming languages. In particular, I enjoy coding in Rust.