Hassan El Mghari

Hassan El Mghari

@nutlope.bsky.social

DevRel Lead at Together AI. Building open source AI apps like roomGPT.io and blinkshot.io. Posting about gen AI & my side projects.

1.2K followers159 following104 posts

Introducing 1kpapers.com! I took the top 1k research papers of the last year, summarized them, and visualized them. Fun fact: all 1,000 papers cost a total of $4 to summarize with DeepSeek V4 Flash.

View full thread

1. Run open source models like DeepSeek, Flux, & Llama across many modalities (chat, image, audio, vision, ect). 2. Finetune models on your own data (small specialized models > big general models). 3. Launch GPU clusters for teams who want H100s / B200s to train their own models.

๐Ÿ“ข Introducing whatisdevrel .com, a resource for developer advocacy! Read to learn about: โ—† How to get a job in DevRel โ—† The three types of DevRel โ—† DevRel advice and resources

Since there's lots of interest, I'm compiling a list of cool startups that are hiring DevRels (mostly US based): - @trychromaย โ€“ hiring manager: @jeffreyhuber - @vercel โ€“ hiring manager: @leeerob - @TimescaleDB โ€“ hiring manager: @avthars - @Get_Writer โ€“ hiring manager: @samjulien

My app BlinkShot just crossed 400k users & 20 million images generated! It has 140k monthly active users who use it to generate images in realtime. Powered by the @togethercompute API โ€“ and 100% free & open source.

2024 Wrapped: โ—† Got married โ—† Got a new job in the AI space โ—† Shipped 13 apps (1M+ users overall) โ—† Created ~10 tutorials (blogs + vids) โ—† Moved to New York โ—† Became a manager Thankful for a lot this year โ€“ย can't wait for 2025!

To celebrate 500k users & 1M apps created, I'm building LlamaCoder 2.0! It'll have a brand new UI, be able to create more impressive apps, and be able to write + execute scripts in TypeScript + Python!

Excited to share we acquired CodeSandBox & are launching a Together Code Interpreter! Been a CodeSandBox user for years so really looking forward to working with the team.

CodeSandbox@codesandbox.io

Today we have some very big news ๐ŸŽ‰ CodeSandbox has joined Together AI and we have joined forces to launch a new product, CodeSandbox SDK! CodeSandbox will continue to run as normal and weโ€™re making private sandboxes free. Check the details below ๐Ÿ‘‡

LLMs are really good at generating code, but running that generated code is difficult. We want to give a way for developers to seamlessly do both. The CodeSandBox SDK solves this! Here's an early preview of LlamaCoder 2.0 generating + executing scripts with the CodeSandBox SDK

I feel like I build too many new surface-level apps + do a bad job of doubling down on apps that do well. In 2025, I'm going to make bigger & better apps, continue to add features to apps that go well, & make sure I make a video + guide for each app I build!

Building an AI wallpaper app! I wrote a script that: 1. Uses an LLM to write 50 good wallpaper prompts 2. Sends them to an image model to generate images 3. Stores the images in a CDN I then manually curate the images by deleting the bad wallpapers & keeping the good ones.

Announcing PicMenu โ€“ a free & open source menu visualizer! It takes a restaurant menu picture and generates nice images for each dish, and works great for lots of non-english menus too. Powered by Llama 3.2 vision + Flux and available today!

I made an app that googles you, shows you a profile of what the internet knows about you, and gives you a "googlability" score! The issue is that it's not always accurate (it still thinks i'm at my last job) so I'm debating whether or not to release it. What do y'all think?

Blown away by how many people have used llama-ocr in 24h! - 6.5k users who uploaded 8k+ images - 4.5k likes on X and 1.4k likes on LinkedIn - 400 GitHub stars & 350+ installs on npm Randomly decided to build + ship in <1 day yesterday. Reminder to just ship things!

Announcing llama-ocr โ€“ a free + open source OCR tool! It takes documents (images for now) & outputs markdown, and does really well for complex receipts, PDFs with tables/charts, ect... Powered by Llama 3.2 vision & available on npm today!

Vision models unlock way better results than traditional OCR and I think we're going to see this method more and more. For now, llama-ocr is a simple wrapper on top of Llama 3.2 vision but I plan to add support for more documents soon like PDFs, JSON output, & more models!

I have so many half built projects like this one, just need to go back and ship them! This one allows people to search up phrases for dev YouTubers and get what video it came from with the timestamp

40 lines is all you need to build a full-stack Next.js AI app with streaming! 5 lines to stream from the backend & 3 lines to read the stream on the frontend with the new Together library ๐Ÿคฏ