About.
I'm a researcher working on generative models, mainly those that can simulate world dynamics (named world models), I'm also interested in what the next generation of generative models will look like, and how to scale them up.
Previously I worked on research in diffusion models with some brilliant friends. I also built a research chat over Twitter known for its high taste, ml_chat. And prior to that I worked on software at several startups (including my own).
I am always cognizant of the fact that I missed out on many possible lives I could have had, and often contemplate what is the best way to increase the likelihood of arriving at the . People coin this “luck surface area”, for me it's more about the unexpected appreciation I have for things I didn't even know I cared for. Perhaps one day I will write about this.
A select timeline of my life:
- 2026
Coming soon
New updates on the horizon... - 2025
500miles on a bike, SFO -> LAX (opens in a new tab)
Took a 500 mile bike ride from San Francisco to Los Angeles, I got a chance to explore the US and see some beautiful scenery. - Apr 2025
San Francisco 🇺🇸 (opens in a new tab)
Moved to the US on O1 visa to get closer to the research community and build with more resources - Jan 2025
Exploring world models @ Lucid.ai
Excited by the prospects of these models and how to scale them up, I started Lucid, got into YC W25, and raised ~$7m from YC and other investors - Nov 2024
Lucid V1 (opens in a new tab)
Released Lucid V1 and it got incredible traction, got featured in major publications and was also used as a baseline by Dreamer V4, this also marked the beginning of my journey into the world of world models. - 2024
Research on diffusion models
Worked on several research projects with some friends, published https://arxiv.org/abs/2412.02352, and started working on video models, unfortunately the project didn't pan out because of NDAs, I pivoted to constraining models to simple environments to focus on the world modelling aspect. - 2024
Made wandview & ml chat
In need of a way to visualize my weights and biases projects I built a mobile app that allows me to do just that, it became incredibly viral and got me a lot of attention (4,000+ researchers used it!), this also led me to start mlchat to learn more about research, it was often referred to as the most productive thing on Twitter. - Jun 2024
Left my startup
The company was severely focused on business dealings and not technical work, so I decided to leave and start working at Toptal as a software engineer that briefly introduced me to ML again on one of my projects - Apr 2021
Cofounded telecom startup
Helped resolve a major bug with .et TLD that caught the attention of the CTO of the largest telecom company in ethiopia, excited by the prospect of building native solutions (current systems are imported from abroad) I started a company, At 18 I won several government telecom contracts that were valued at M$+.
Fun fact, My company helped launch spotify in Ethiopia :) - Jan 2019 — Dec 2021
360Ground
Started helping them with Google natural machine translation models and ended up taking several roles including design and development, I ended up becoming the CTO/Tech lead briefly after the existing CTO was poached by microsoft, there, I worked on one of the largest apps in ethiopia at the time... - Mar 2017
Dropped out of junior high school to pursue my love for building things, interned at iCog Labs (known for Sophia realistic humanoid)
I've always learned through experiments and association, building intuitions before formalising them. School wanted the opposite: absorb the conclusion, then move on. That mismatch meant I was always doing twice the work to arrive at the same place my peers started from. Early in high school, I realised I'd learn faster outside of it. The only robotics lab in my city took me in, and that was reason enough to drop out. - ~ 2014
Started coding at 11 years old
Built a website for my school to help them manage assignments and attendance, it was perhaps the first time i felt like I was doing something meaningful