Hassan El Mghari

Hassan El Mghari leads developer experience at Together AI and builds open-source AI products used by millions of people. His best-known projects—RoomGPT, RestorePhotos, and LlamaCoder—turn advances in image generation, photo restoration, and coding models into accessible applications that other developers can study and adapt.

From developer advocate to AI product builder

El Mghari began learning to code in 2020, switched his college concentration from electrical to computer engineering, and graduated in 2021. After an early developer-advocacy job ended, he joined Vercel that September and became a senior developer advocate the following year. Following a brief period consulting on AI applications, he joined Together AI in January 2024, initially leading developer relations before expanding into developer experience and building a team around documentation, demonstrations, and developer education.

His first consequential AI experiment automated accessibility descriptions for hundreds of conference photographs. Other early projects included AI Commits, a command-line tool that converts Git diffs into commit messages, and RestorePhotos, which uses GFPGAN to sharpen faces in old photographs. His RoomGPT project applied ControlNet to interior redesign, preserving a photographed room’s structure while generating new decorative styles.

At Together AI, LlamaCoder combined Llama 3.1 405B with hosted inference to turn written requests into working applications with browser previews. By September 2024, it had generated more than 200,000 applications. His portfolio also includes BlinkShot for real-time image generation, Self.so for personal websites, and LogoCreator for brand imagery.

• Simple applications, immediate utility. El Mghari favors products with understandable purposes, minimal interfaces, and often only one or two model calls. He recommends launching quickly with existing models, validating demand before adding complexity, and updating applications as better models appear.

• Open-source distribution and practical economics. Publishing complete implementations helps developers learn, contribute, and share useful products. El Mghari supports free access through usage limits and partnerships with infrastructure providers that sponsor inference, databases, authentication, or other services.

• Design taste in AI-generated software. He treats mobile usability, layout, loading states, animation, and visual clarity as decisive product features. His Hallmark design skill codifies common generated-interface mistakes and combines explicit constraints with visual references and reusable themes. He recommends preserving preferences in AGENTS.md, providing screenshots and detailed context, and refining initial outputs through repeated iterations.

• Design variation and economical iteration. His /variate announcement introduced an open-source skill that generates multiple interface variations for Claude Code, Codex, and OpenCode. He also evaluates less expensive open models against specific design tasks, using faster iteration to improve interfaces without assuming higher prices guarantee better results.