hi, i'm gabe

welcome to my little corner of the internet

i'm a software engineer based in seattle, wa working at DoorDash studying ai safety and building fun stuff in my freetime

if you want to contact me - gab.01@hotmail.com
GitHub | LinkedIn


check out my latest project, pancake, a parallel agent manager

GitHub | LinkedIn/Demo

blog

see all blog posts →

about me

My name is Gabriel Mousa, I'm a full-time software engineer, and I have fun working on technical things. Currently, my interests lie in ai safety with a particular focus on LLM evals, supervised-fine tuning, rlhf, and any unique ways I can potentially improve upon the post-training process to make models safer for the world. As a whole, I believe AI is inevitable and while there are incredible benefits resulting from it, safety needs to be a core tenant for those building it in order for society to build competently.


Outside of work I'm an avid sailor, cook, swimmer, rock-climber, hiker, and traveller. I live at home with two wonderful cats and most important of all an even more wonderful partner who supports me in all of my endless endeavors. I like to read and try my best to take every moment as a learning opportunity, while leading with curiosity.

resume

want to know more about my work?

download resume (pdf)

ideas

random thoughts and projects i'm thinking about

subset of model weights unique to inference layer, to allow for model adaptability
deception llm evals
sailing the Race To Alaska
building beneficial evals for llms *credit to adam jones*
programming language for llms thats highly optimized
generating synthetic data using llms as a judge with a set of evals for new data before acceptance
security screener for new ai projects to make startup simpler and safer
pancake a parallel agent manager and how to improve upon it