Asanshay Gupta
About
I'm an engineer, researcher, and designer. I just graduated from Stanford, and now I'm building something new. Recently, I've been interested in continual learning and proactivity, but I've worked extensively in GPU systems, human-computer interaction, and robotics. Outside of work, I mentor FRC 1414, do some brand design work, and build a ton of projects, which you can explore below.
Python and Typescript packages, coding agent, and RLHF using in context reinforcement learning.
A framework for generating and orchestrating an ever-evolving library of stateful, specialized agents.
Training agents in parallel using an evolutionary PBT on top of PPO in the Madrona batch simulator.
Bare metal driver for the Raspberry Pi's GPU with over 50% of the code written in assembly including hardware profiling.
A real-time AI agent platform for market research and analysis. Won 1st place at CMU's AI Valley Hackathon.
A Python library for generating academic citations from URLs using scrapers. Used by 2000+ researchers.
Dynamically create sub-agents on the fly to solve multifaceted tasks. SOTA in complex task benchmarks.
Automates academic writing using a hierarchical multi-agent system with both deep and wide research.
Studied the human interaction of dropping into an office by building a virtual office system for Stanford's Gates Computer Science building.
Python and Typescript packages, coding agent, and RLHF using in context reinforcement learning.
Bare metal driver for the Raspberry Pi's GPU with over 50% of the code written in assembly including hardware profiling.
A real-time AI agent platform for market research and analysis. Won 1st place at CMU's AI Valley Hackathon.
A Python library for generating academic citations from URLs using scrapers. Used by 2000+ researchers.
Automates academic writing using a hierarchical multi-agent system with both deep and wide research.
A framework for generating and orchestrating an ever-evolving library of stateful, specialized agents.
Training agents in parallel using an evolutionary PBT on top of PPO in the Madrona batch simulator.
Dynamically create sub-agents on the fly to solve multifaceted tasks. SOTA in complex task benchmarks.
Python and Typescript packages, coding agent, and RLHF using in context reinforcement learning.
Dynamically create sub-agents on the fly to solve multifaceted tasks. SOTA in complex task benchmarks.
Training agents in parallel using an evolutionary PBT on top of PPO in the Madrona batch simulator.
A real-time AI agent platform for market research and analysis. Won 1st place at CMU's AI Valley Hackathon.
A framework for generating and orchestrating an ever-evolving library of stateful, specialized agents.
Bare metal driver for the Raspberry Pi's GPU with over 50% of the code written in assembly including hardware profiling.
A Python library for generating academic citations from URLs using scrapers. Used by 2000+ researchers.
















