Prannay Hebbar

AI Researcher

Building cutting-edge AI systems at AGI Inc. Published researcher in web agents and reinforcement learning. Stanford alumnus passionate about advancing artificial general intelligence.

Learn More

About Me

I'm an AI researcher at AGI Inc, where I build cutting-edge AI systems and advance the field of reinforcement learning. I studied at Stanford University under Prof. Stephen Boyd, focusing on defining convex constraints on RL problems.

My work includes publishing research on web agents benchmarking, creating evaluation frameworks, and implementing state-of-the-art RL algorithms. I'm passionate about training RL on games and simulations, and I believe strongly in the potential of artificial general intelligence.

12K+

Package Downloads

$500K

Contract Closed

76%

Web Agent Reliability

Technical Skills

Programming

Python C++ JavaScript Java Bash

Frameworks

PyTorch Ray Docker LangGraph ReactJS NodeJS

Experience

Mar 2025 - Present

AI Researcher

AGI Inc. - San Francisco, CA

Published paper on REAL: Sandbox Websites Agent Benchmarking, securing $500K Amazon Nova contract
Created and maintain agisdk with 12K+ downloads for web agent benchmarking
Improved web agent reliability to 76% using policy distillation and VerL/GRPO implementation
Implemented Torch profiling on NVIDIA H200 GPU cluster with W&B logging
Built evaluation dashboard for reinforcement learning training

Jan 2023 - Mar 2025

System Analyst

Aethereus (acq. Myridius) - Dallas, TX

Bread Financials: Implemented Comenity Bank Cardholder Portal across 46+ partner brands
Digitized partner-brand onboarding, reducing integration cycle time by 10-15 days
Launched omni-channel chat portal handling 1.4K messages/week with <100ms latency
Migrated 9 critical APIs from MuleSoft to Azure, reducing defects by 20%
California Lawyers for the Arts: Automated data interchange between SaaS platforms
YellowPad AI: Built RAG pipeline achieving 95% clause recall in legal contracts

Featured Projects

Claude-Web

Enables Claude code to act as a web agent, providing autonomous web interaction capabilities.

Python Web Agents AI

Mac Control

Hackathon-winning project using MCPs to control Mac actions through AI interfaces.

MacOS MCP Automation

agisdk

Web agent and harness providing sandbox website access and benchmarking for major AI models.

Python Benchmarking 12K+ Downloads

Publications & Media

REAL: Sandbox Websites Agent Benchmarking

arXiv preprint - Research paper on benchmarking web agents in sandbox environments

Demo Videos

Technical demonstrations of AI systems and web agent capabilities

Education

Stanford University

Jun 2024 - Sep 2024

Visiting Student - GPA: 3.775/4

Courses: HPC, Convex Optimization, Principles of Robotics, AI Principles & Techniques, Investment Science

Research under Prof. Stephen Boyd on convex constraints in RL problems

VIT University

Jul 2019 - Apr 2023

B.Tech Computer Science and Business Systems - GPA: 3.26/4

Get In Touch

Let's connect!

I'm always interested in discussing AI research, collaboration opportunities, or innovative projects.

hebbarpran@gmail.com

+1 (209) 630 9035

San Francisco, CA