Sidharth Kannan
I am a Research Engineer at ChipAgents, working on building more efficient agents for electronic design automation. Previously, I studied Physics and Computer Science at UC Santa Barbara.
Research and writing
Selected work
Λ-conditional flow matching: scale-aware representation learning for cosmology with flow matching
Unsupervised flow matching learns compact, scale-aware representations of cosmological simulations for reconstruction, synthetic data generation, and parameter estimation.
Graph Neural Networks for Interferometer Simulations
Graph neural network surrogates that reproduce LIGO interferometer simulations up to 815 times faster, supporting more efficient instrumentation design.
Build-Your-Own-LM: A Journey into Language Modeling
A from-scratch exploration of tokenization, transformer architecture, caching, quantization, and modern language-model components.