Prashant Pandey
Assistant Professor, Khoury College of Computer Sciences, Northeastern University, Boston.
p.pandey@northeastern.edu
440 Huntington Avenue
Boston, MA 02115
I am an Assistant Professor in the Khoury College of Computer Sciences Northeastern University, Boston, where I direct the SALT Systems Lab and co-direct the Data Lab. I am also an Affiliate Faculty in the Performance and Algorithms Research Group at Berkeley Lab.
My goal as a researcher is to build scalable data systems with strong theoretical guarantees and employ them to democratize next-generation data analyses. My work combines filters and compact data structures, I/O-efficient algorithms, and high-performance and GPU computing, applied across computational biology, stream processing, and storage.
Prospective PhD students (Fall 2027): I am recruiting 1–2 PhD students to work on large-scale GPU computing and AI systems. Projects span four areas: data structures, especially adaptive filters (Aeris Filter, SIGMOD ‘26; To Adapt or Not to Adapt, Best of SIGMOD ‘26; ThimbleQF, SIGMOD ‘27); GPU computing, including memory management, scheduling, and concurrent data structures (WarpSpeed, ALENEX ‘26); vector search (Jasper, VLDB ‘26; Directional Beam Search, VecDB ‘26); and AI systems, covering LLM inference (Iceberg Attention, InferScale) and GPU-scale indexing of LLM training corpora (Cascade).
I am looking for students with a strong background in one or more of: data structures and algorithms, GPU programming (CUDA), LLM inference, and database systems.
If you are interested, please mention my name in your PhD application to Northeastern.
Previously, I was a Research Scientist at VMware Research for a year. Prior to that, I did postdocs at UC Berkeley/LBL and Carnegie Mellon University.
I love outdoors and occasionally like to scribble my experience in a blog.
news
| Oct 14, 2026 | New preprint: Iceberg Attention: Sublinear Long-Context Decoding with Mixture-of-Gaussians. At 1M tokens, MoG decoding is 27× faster than vLLM with FlashAttention, while matching exact attention and reading less than 5.3% of the KV cache. |
|---|---|
| Jul 24, 2026 | Our paper To Adapt or Not to Adapt, That is the Ski Question was selected as Best of SIGMOD 2026! |
| Jul 17, 2026 | Hunter McCoy successfully defended his Ph.D. thesis! |
| Jul 02, 2026 | GPU-Accelerated ANNS: Quantized for Speed, Built for Change (Jasper) was accepted to VLDB 2026! |
| Jun 01, 2026 | Three papers accepted to SIGMOD 2026: Aeris Filter, Breadcrumb Filters, and To Adapt or Not to Adapt, That is the Ski Question. |
selected publications
- SIGMOD 2026To Adapt or Not to Adapt, That is the Ski QuestionIn ACM Special Interest Group on Management of Data 2026, 2026