Sharif Shameem

Sharif Shameem is a software developer and entrepreneur based in San Francisco who has founded three AI-related startups: Vectordash, a peer-to-peer GPU cloud that went through Y Combinator in 2019; Debuild, a GPT-3-powered code generation tool whose demos went viral in July 2020; and Lexica, an AI image search engine and generation platform backed by Founders, Inc..¹⁵ In January 2026, Shameem joined OpenAI to lead OAI Labs, a research group focused on prototyping new interfaces for human-AI collaboration.⁷

Early life and education

Shameem studied computer science at the University of Maryland but did not graduate.¹¹ In a 2023 interview, he said the decision to drop out was influenced by a conversation with an investor who had also left college for Y Combinator and later returned, framing the choice as reversible, and by reading Bryan Caplan's book The Case Against Education.² Before founding his own companies, Shameem worked at MITRE, a federally funded research and development center, where he did computer vision research.²

Vectordash

Shameem's first startup grew out of a trade with a friend who was mining Ethereum on an NVIDIA 1080 Ti around 2016-2017.² Shameem wanted to train a character-level recurrent neural network on his iMessage history but needed GPU access; renting a T4 on AWS cost $60-80 and was slower than consumer hardware.² He paid his friend $10 per day for GPU time, a rate that exceeded Ethereum mining revenue, and the arbitrage became the basis for Vectordash: a marketplace connecting cryptocurrency miners' idle GPUs with machine learning researchers.²

The platform launched out of Y Combinator's Winter 2019 batch, accumulating over 120,000 NVIDIA GPUs and offering rentals at roughly 20 cents per hour.²⁴ Vectordash charged customers $28 per month for cloud gaming, streaming games from GPUs located within approximately 300 miles of the user to keep latency under 20-30 milliseconds.⁴ Shameem demonstrated Apex Legends running at roughly 100 frames per second on a 13-inch MacBook.⁴ The company's distributed model predated Google Stadia's launch by about a year.²

Vectordash struggled to convert its initial user base of college students and researchers into paying customers, and larger companies were reluctant to trust a startup with their data.²

Debuild

In July 2020, Shameem gained access to OpenAI's GPT-3 API after tweeting a request for beta access; Greg Brockman, then OpenAI's president, responded the same day.² Shameem built a tool that accepted natural-language descriptions and generated JSX React components, compiling and rendering them directly in the browser.² On July 13, 2020, he posted a 30-second demo to Twitter showing a layout generator that converted text descriptions into working UI code.⁹ The tweet received over 20,000 likes, and Shameem spent the following week releasing additional demos, including generating a functioning React application from a text prompt.²

The project became Debuild, a no-code tool for building web applications by describing their appearance and behavior in plain English.³ Shameem trained the system on code datasets from sources including GitHub and Stack Overflow.³ In a September 2020 interview with Business Insider, he said the company had fewer than five employees and predicted that GPT-3's code-writing capabilities would "increase the number of programmers in the world" rather than replace them.³ WIRED described the tool as offering "a text-to-code tool for building web applications."¹²

Lexica

Shameem launched Lexica on August 23, 2022, one day after Stability AI open-sourced the Stable Diffusion model weights.² He had been spending hours in the Stable Diffusion Discord server watching users generate images with complex prompts but found the Discord search bar inadequate for browsing them.² Over roughly two days, he scraped approximately 10 million images from the Discord, built a search interface using PostgreSQL full-text search, and shipped the site.²

Within 24 hours of launch, Lexica processed 51,000 search queries and served 2.2 terabytes of bandwidth.² By the second day, those numbers doubled to 111,000 queries and 4.5 terabytes, with 22 million images served.² The platform reached one million users within approximately ten days, the threshold Shameem had set for treating the project as a full-time company.²

Shameem later added semantic search powered by CLIP embeddings and Facebook's FAISS library for fast nearest-neighbor lookups, enabling users to search by image content rather than prompt text alone.² The system ran on CPUs for serving, with GPUs used only to compute the embeddings.²

Funding and team

In September 2022, Lexica raised a $5 million seed round led by Daniel Gross and Nat Friedman through AI Grant.⁶ Shameem announced the round on September 28, writing that Lexica's goal was "to give everyone the ability to create whatever they imagine."⁶ Lexica was co-founded by Shameem, Dylan Molina, and Subhash Ramesh, and as of 2026 had a team of four.⁵

Aperture

In December 2022, Shameem introduced Lexica Aperture, a proprietary text-to-image model developed in-house.¹⁰ The first version produced photorealistic images from text prompts, and a second version followed later that month.² Aperture v3 shipped in April 2023.² As of May 2023, Lexica was serving over 5 billion images per month.² The platform is in the Founders, Inc. portfolio.⁵

OpenAI

On January 5, 2026, Shameem announced on X that he had joined OpenAI, writing: "there are a few rare inflection points throughout history" in which new interaction paradigms for computing are established.⁷ He described his role as leading OAI Labs, "a research-driven group focused on inventing and prototyping new interfaces for how people collaborate with AI."⁷ TrendForce reported that Shameem officially joined on January 6, 2026.⁸