The site brings together his blog, research interests, project updates, and worklogs, offering readers a look at both his technical work and his informal reflections. His current focus includes machine learning systems, learning CUDA, writing GPU kernels, and studying technical reports from recent open-source model releases. He also takes part in weekly hackathons and machine learning events in Tokyo, using them to build projects and share what he learns.

Sumit is particularly interested in large language model training, inference optimization, fine-tuning, and research on low-resource languages. Topics he explores include dataset curation, tokenization, cross-lingual transfer, and adapting models for underrepresented languages. The site is also an invitation to connect about mentorship and collaboration. Sumit is open to compelling deep learning problems beyond his current focus and values hands-on learning alongside experienced researchers. Visitors can follow his writing and project progress or contact him by email or through social media.