Hetansh Waghela

Welcome!

I am a B.Tech student in Computer Science and Engineering (Data Science) at Dwarkadas J. Sanghvi College of Engineering (CGPA 9.56/10), focused on understanding how language models represent and process information internally.

My research centers on mechanistic interpretability and machine unlearning — probing whether unlearning methods truly erase knowledge or merely suppress it, whether sparse autoencoder features transfer across model architectures, and where post-trained models learn to conceal affect. I am currently a Research Intern at Georgia Institute of Technology under Prof. Jing-Yung Chen, working on glacier-flow dynamics through reproducible data-analysis and modelling workflows. I am also a Forward Deployed Engineer Intern at X-ARC, building and hardening production agentic systems with tool use, retrieval, memory, and multi-step orchestration.

Previously, I was an AI Engineer Intern at AhumLabs, where I built retrieval and assistant infrastructure for a multi-tenant trading platform — including tenant-specific knowledge ingestion, transaction indexing, and admin tooling for onboarding and form customization.

Outside internships, I build research and engineering projects across interpretability (Cross-Model SAE Feature Transfer, Subliminal Unlearning, Lazarus, Ghosts in the Model), biomedical AI (ImmunoPath), and compliance-grade systems (ANVIL). I am an open-source contributor to Stanford CRFM HELM and TransformerLens. I like projects that combine rigorous evaluation with systems that people can actually use.