Charan Somalraju

Handshake Ai Fellow @Handshake

New York, NY, US

Working on Canary, an advanced evaluation project aimed at identifying reasoning failures in large language models (LLMs)- Designed and executed high-level prompt-based tasks (image + text) to “stump” models by exposing weaknesses in reasoning and interpretation- Developed image-dependent prompts that required the model to extract structured information (e.g, graphs, architectures, numerical data) and perform logical or mathematical reasoning- Analyzed model outputs to identify reasoning inconsistencies or hallucinations, then refined prompts or corrected reasoning to improve reliability.