Deshraj Yadav

Deshraj Yadav is co-founder and chief technology officer of Mem0, a San Francisco, California, company that builds a memory layer for AI agents, so that apps built on large language models can remember a user's preferences and past conversations from one session to the next. He started it in 2023 with Taranjeet Singh, a friend since their undergraduate days in India.

Open-Source Research in India and Virginia

Yadav studied computer science and engineering at JSS Academy of Technical Education in Noida, India, where he met Singh in 2012. As an undergraduate he worked on CloudCV, an open-source computer vision project, through Google Summer of Code in 2015, and he later served as one of its organization administrators. He then spent a year as a research assistant in Virginia Tech's machine learning and perception lab, working on Visual Dialog, research on AI agents that hold a conversation about an image, which was a spotlight paper at CVPR 2017, according to his site.

Building EvalAI at Georgia Tech

At the Georgia Institute of Technology, where he earned a master's degree in computer science, he built EvalAI, an open-source platform for evaluating and comparing machine learning algorithms at scale, as his thesis, which he defended in December 2018. He and Singh had built it together and grew it to about 1,600 GitHub stars, TechCrunch reported. He won a Snap research scholarship in 2017, one of eight awarded worldwide, and interned at Snap in 2018. In 2019 he co-founded Caliper, which let recruiters test practical AI skills through challenges based on real problems.

Leading the AI Platform at Tesla Autopilot

Yadav joined Tesla Autopilot in July 2019 and became a senior engineer leading its AI platform, which tracked large training runs and model evaluations and monitored jobs and problems on the training cluster. Before that he worked on Autopilot's AI tooling, including a 3D labeling platform, model versioning and active learning to improve labels.

Giving AI Agents a Memory

When Singh moved to the United States, the two decided to build something together. A meditation app they made in late 2023, based on the teachings of the Indian yogi Sadhguru, went viral in India, but users complained that it did not remember anything about their meditative journey. That pushed them to turn Embedchain, Singh's open-source project for retrieving data, into Mem0. On Y Combinator's Founder Firesides, Yadav explained that Mem0 sorts incoming information into key-value pairs, semantic chunks or a graph of related facts, then pulls from all three in real time. That spares developers from stuffing everything into the model's context window, which costs more and runs slower.

His advice to other founders is about focus, borrowed from computer science. When starting a company, "you want to make sure you apply DFS rather than applying BFS so that you stay focused," he said, meaning depth-first rather than breadth-first search.

Personal Life

Outside work, Yadav plays cricket in two leagues in the San Francisco Bay Area.