Hammad Rizwan’s research focuses on process-level understanding and control of LLMs: pruning-based circuit discovery, range-based neuron analysis to localize concepts, and disentangled representations for precise, low-interference edits. Emphasis on scalable methods that preserve performance and reduce bias.