Sabyasachi Basu

I am a postdoctoral researcher at Microsoft Research, in Bangalore, India. I work in the DiskANN team, under Ravishankar Krishnaswamy. I am currently thinking about problems on quantization, matryoshka embeddings, and OOD queries. 

In September 2025, I received my PhD from the University of California, Santa Cruz, under the supervision of C. Seshadhri. My thesis was titled "Decomposition Techniques for Web-Scale Networks"; please go through my unnecessarily long acknowledgements that took me as much time to write as the rest of my thesis (/s). Before UCSC, I was an undergraduate at the Indian Institute of Science, Bengaluru, majoring in math. My advisors there were Vishwesha Guttal and Srikanth Iyer. I can send you a CV upon request, but most of the relevant stuff is on my website.

In the summer of 2023, I was a student researcher at Google, working with Aneesh Sharma to understand retrieval in graph embeddings and designing better training/retrieval procedures to obtain better results. In 2022, I worked in the personalization team at Walmart, and in 2020, I interned at the Data Science Institute at Lawrence Livermore National Laboratory.

Till recently, my Erdős number was 3. If preprints count, it is now a 2. 

Research Interests

Broadly, I am interested in building algorithmic techniques that have a strong theoretical foundation, but can also be implemented efficiently in practice. Most modern computer science use cases require data; usually copious amounts of it. I like to think that the best tactic for algorithm design in this day and age is understanding the empirical properties of the data, and then building theory around it. I think of this as more of a physics-like approach to CS, rather than the more prevalent model-first approach to theory. I think it is especially crucial for those who wish to build methods to bridge theory and practice.