About

Pratyush Kerhalkar

Contact: pratyushkerhalkar@gmail.com | +1 (602) 323-4896

I am a Computer Science Master's candidate at Arizona State University with 3+ years of applied experience in machine learning, backend engineering, and data architecture.

Professional Journey

Currently, I work as a Research Student at ASU's Interactive Robotics Lab, where I work on LLM-based optimization and shared-control robotics for manipulation.
Previously, in Deloitte's AI & Data vertical, I built high throughput backend systems to process financial records.
Earlier, as an R&D Student at Samsung Research Institute Bangalore, I developed NLP-based low-latency microservices to identify target devices for Samsung Bixby.

View detailed experience here.

Research Interests

Apart from production software engineering, I am deeply involved in machine learning research, with a focus on the interests listed below.

Meta Learning
Optimization
Large Language Models
Reinforcement Learning
Multimodal Systems

News

Nov 27, 2025
I am super excited to share that I will be attending NeurIPS 2025 and presenting our work on policy optimization via LLMs. Looking forward to meeting other researchers and discussing new ideas.
Oct 12, 2025
Thrilled to present our work on program synthesis for robot policy optimization at ASU's SouthWest Robotics Symposium.
Aug 22, 2024
Excited to begin my Master's in Computer Science at Arizona State University and start this next chapter in machine learning, systems, and research.

Selected Projects

ProPS: Prompted Policy Search

Co-developed an LLM-based optimizer that introduces linguistic reasoning into numerical and reinforcement learning policy optimization, outperforming PPO and TRPO on multiple benchmarks.

Paper Code Tutorial

Shared Control UR5 for In-Space Welding

Designed a human-in-the-loop orbital construction setup with marker-based pose estimation and semi-autonomous Grab and Weld modes for stable, precise end-effector alignment.

Article Poster

Select Publications

Prompted Policy Search: Reinforcement Learning through Linguistic and Numerical Reasoning in LLMs - Paper
EinsteinPy: A Community Python Package for General Relativity - Paper
Cross-Geography Generalization for Classification of Flooded Regions in Aerial Images - Paper

Pratyush Kerhalkar

Pratyush Kerhalkar

News

Selected Projects

ProPS: Prompted Policy Search

Shared Control UR5 for In-Space Welding

Select Publications

Explore in Detail

Professional Experience

Projects

Research

Achievements & Activities