Jiwon Jeong, Dun Li Chan, Andrew Ceniccola, Rahul Sharma, Simon Sang, Parva Mehta, Pietro Di Nicco, Karim Saraipour, Marko Karbevski
SLM Agents Workshop, NeurIPS 2026
Argues that linear query transformations limit the expressivity of attention, and makes the case for nonlinear query mappings in small language models. Shows that nonlinear queries can recover capabilities that purely linear attention heads cannot represent.
Rahul Sharma
Second Workshop on Secure and Trustworthy LLMs (SeT-LLM), 32nd ACM SIGKDD · Jeju
Studies whether linear probes of context-trust in RAG models yield usable control directions. Finds that directions which are readable by probes often fail to steer model behavior reliably, cautioning against using probe-derived directions as control surfaces.
Bhuvan Sachdeva, Pragnya Ramjee, Rahul Sharma, Mithun Thulasidas, Sowmya Raveendra Murthy, Geeta Fulari, Kaushik Murali, Mohit Jain
European Journal of Ophthalmology
Evaluates an LLM-powered WhatsApp chatbot with doctors in the loop for cataract surgery patients. Reports on its utility for pre- and post-operative care based on a deployment across multiple hospitals.
Rahul Sharma, Pragnya Ramjee, Kaushik Murali, Mohit Jain
GenAI for Health Workshop, NeurIPS 2025 · San Diego
Analyzes how doctors edit LLM-generated answers to patient questions in an expert-in-the-loop health chatbot. Characterizes edit patterns and draws design implications for human-AI collaboration in healthcare.
Rahul Sharma, Harshit Sharma, Tamanna Meena, Padmavati Khandnor, Palak Bansal, Paras Sharma
6th IAPR Conference on CVIP · IIT Ropar
Benchmarks deep learning object detectors for ship detection in satellite imagery. Compares accuracy and speed trade-offs across model architectures.