Engineer & ML researcher · building evsys-sdk
Hi, I'm Shrey. I work on post-training and continual learning for language models, and I'm building evsys-sdk, an open-source SDK for exactly that (docs). I studied Computer Science at IIT Bombay and completed an MS in Computer Science at the University of Chicago.
I have a background in ML research with papers at ICLR and NeurIPS, and I like building things at the early, messy stage of a company. Outside of work it's long runs, lifting, and a lot of rock and rap.
Why continual learning is better seen as a closed-loop system with autoresearch inside it, rather than a single learning algorithm.
Optimizing federated learning through Bayesian coresets.
Speeding up malware classification with a Vision Transformer model.
A selective uncertainty propagation mechanism to improve efficiency in offline reinforcement learning.
More on Google Scholar.
Built automation tools for startup sourcing and ran AI market research.
Built a CSV editor to automate reliability-testing workflows.
Worked on voice authentication systems and backend, deployed on Google Cloud.
Built an OCR-based discrepancy checker for trade-document validation.
Graph Neural Networks over system audit logs to flag potential security threats.
Efficient video summarization by clustering frames with LSH.
A sequence transformer with a BERT encoder and GPT-2-style decoder for autoregressive pixel prediction.
State-tied triphone HMMs for Wolof and Marathi using Kaldi and SoX.
A peer-to-peer file-sharing network over TCP sockets, using BFS for search.