Hassan El Mghari

Hassan El Mghari, known online as Nutlope, builds open-source AI applications and teaches developers how to build with AI. His personal website identifies his current role as leading Developer Experience at Together AI, where he manages a team working on demos, documentation, and content.[2][3]

Among his projects is LlamaCoder, an open-source application for generating small apps from a single prompt. Together AI developed the application; El Mghari hosts its repository under his Nutlope account and discussed its use in a September 2024 Meta feature. He described people making quiz apps, pomodoro timers, and budgeting apps through prompting, and judged Llama 3.1 405B to be the first open-source model that performed well for this particular coding use case.[2][4][6]

LlamaCoder’s current documentation specifies Llama 3.1 405B for generation and Together AI for inference. Its browser preview uses esbuild-wasm and esm.sh and runs inside a sandboxed iframe, allowing users to see the generated application.[4]

In the September 18, 2024 Meta feature, El Mghari argued for open-source AI on the grounds of ownership and flexibility: “it enables organizations to truly own their data, their models, and have that full flexibility instead of getting locked in.” Meta reported that LlamaCoder had generated more than 200,000 apps just over a month after release—a count of generated apps, rather than unique users.[6]

Before his Together AI work, El Mghari was Senior DevRel at Vercel. His January 2023 account of building RestorePhotos provides an earlier example of his approach to combining AI APIs, accessible interfaces, and public teaching.[5]

El Mghari built RestorePhotos after discovering an image-restoration API on Replicate and testing a prototype on an old photograph of his father doing karate. In his January 18, 2023 retrospective, he recalled taking about a day and a half over a weekend to complete the application. Before launching it on January 9, he gauged interest with a screenshot, polished the interface, and incorporated feedback from Guillermo and Malte at Vercel.[5]

The original application used a Next.js frontend, upload.io for photo storage, and a Next.js serverless backend. The backend submitted the uploaded image to Replicate, polled until processing finished, and returned the restored image to the frontend. El Mghari reported 53,000 unique users and 79,000 restored images after nine days, alongside 139,000 page views.[5]

Keeping RestorePhotos free created an inference-cost problem. El Mghari reported receiving a $240 Replicate bill and securing sponsorship from Replicate in exchange for credit in the footer and README. He said he chose learning, open sourcing, and inspiring others over charging for the application, while also building an audience for possible paid products later.[5]

His lesson from RestorePhotos and ten side projects shipped in 2022 was that “simple products win.” He favored products that did one thing well, had a good interface, and were free or easy to start using. In January 2023, he announced plans to concentrate almost exclusively on AI side projects, describing himself as a full-stack developer without a machine-learning background who enjoyed using APIs creatively and teaching others how the resulting applications worked.[5]