Vedaant Jain

Hey, I am Vedaant, Math&CS major at UIUC. As an undergrad, I have worked on various machine learning problems.

Intially, I designed rigorous evaluations, HumorDB, to benchmark complex scene reasoning in vision and multimodal models. Then I helped develop the first certification framework for LLM’s reading comprehension, LLMCert-C, to move beyond static test sets and formalize desired properties. Nowadays, I focus on explainable adversarial scene generation for multimodal models and ML for biology.

Additionally, on the systems side, I have written custom CUDA and AWS Trainium kernels, and implemented a unified paging system for KV cache and multi-rank LoRA adapters, similar to S-LoRA.

I also interned at Kumo.AI in summer of 2025 and Metaphor Data in summer of 2024.

Some of my fun class projects have included: Interpretability for priming/prefilling attacks, mimicking reddit users with LLMs, curriculum learning/RL for embodied agents, parkinson’s disease progression.

My email: vvjain3@illinoi.edu, Github: https://github.com/Vedaant-J