Location: DC/NoVA
Remote: preferrable, but willing to come in one or two days a week
Willing to relocate: unlikely, but possible
Technologies: C, C++, Python, x86 & ARM assembly, Altium/Kicad, JavaScript, HTML/CSS
Résumé/CV: ask me
Email: xevek@protonmail.com
i'm a reverse engineer, software developer, and embedded systems expert with about 15 years of experience. my secret to avoiding burnout is to work on and learn as many new things as possible in full immersion style and i'm currently in the throes of a raging ML obsession -- i would love to do more work related to it.
as far as hardware goes, i'm experienced with robotics (remote control, sensors, motors, actuators, etc), motion simulators, microcontrollers, embedded x86, audio amplifiers, express racks (for the ISS) and a bit of RF and analog electronics. on the software side, i'm intimately familiar with c, c++ and python, but i can read most code and understand it, including assembly (there were a lot of misspent hours in IDA pro during my youth). while i know javascript, html, and css i am not interested in writing web apps again.
when it comes to ML, i'm proficient with pytorch debugging and i can full finetune extremely large models (120B+) with sft or offline RL (dpo/kto). i've had to do my own data wrangling, so i'm familiar with most of the chat templates/prompt formats of LLMs and converting to and from the various dataset formats that training software uses. i am familiar with most LLM architecture internals and i follow mechanistic interp papers and research, so i train a lot of steering/control vectors on gemma and peek at them with the gemmascope sparse autoencoders. i've trained new output heads for pretrained models, permanently imbued models with control vectors and run abliteration/orthogonalize activation steering, and i can fix your axolotl container when it inevitably breaks under the weight of the hf/torch ecosystem.
i'm a seasoned hacker but my position has turned into project management, which i hate.
tech lead or bust. liberate me and i'll be your friend forever.
Do: contact me if you're looking for a professional who wants to get things done.
Don't: contact me if you want me to sit in meetings all day. i am not built for that.
as far as hardware goes, i'm experienced with robotics (remote control, sensors, motors, actuators, etc), motion simulators, microcontrollers, embedded x86, audio amplifiers, express racks (for the ISS) and a bit of RF and analog electronics. on the software side, i'm intimately familiar with c, c++ and python, but i can read most code and understand it, including assembly (there were a lot of misspent hours in IDA pro during my youth). while i know javascript, html, and css i am not interested in writing web apps again.
when it comes to ML, i'm proficient with pytorch debugging and i can full finetune extremely large models (120B+) with sft or offline RL (dpo/kto). i've had to do my own data wrangling, so i'm familiar with most of the chat templates/prompt formats of LLMs and converting to and from the various dataset formats that training software uses. i am familiar with most LLM architecture internals and i follow mechanistic interp papers and research, so i train a lot of steering/control vectors on gemma and peek at them with the gemmascope sparse autoencoders. i've trained new output heads for pretrained models, permanently imbued models with control vectors and run abliteration/orthogonalize activation steering, and i can fix your axolotl container when it inevitably breaks under the weight of the hf/torch ecosystem.
i'm a seasoned hacker but my position has turned into project management, which i hate. tech lead or bust. liberate me and i'll be your friend forever.
Do: contact me if you're looking for a professional who wants to get things done. Don't: contact me if you want me to sit in meetings all day. i am not built for that.