Member of Technical Staff, Post-Training
Automate your job search with Sonara.
Submit 10x as many applications with less effort than one manual application.1
Reclaim your time by letting our AI handle the grunt work of job searching.
We continuously scan millions of openings to find your top matches.
Overview
Job Description
About Us
Chakra Labs' mission is to encode human taste into intelligence. We build high-fidelity trajectories and environments for frontier AI research, working with several of the top labs.
Why you should work with us
You will get to touch the latest and greatest technologies across the data, AI, and infrastructure stack.
You will get mentored by an experienced and qualified team.
You will get paid a very competitive cash package with top benefits and a $30 daily meal stipend. We also provide a compelling equity package.
Why you shouldn’t work with us
This is our life’s work. We have fun, but we work hard. If you want to clock a 9-5, this probably isn’t a great fit.
There’s no clear playbook for success. Data and AI are constantly shifting. While we have a clear mission and vision, if you’re uncomfortable with ambiguity this role will be challenging for you.
About You
You have a Master / PhD level knowledge of machine learning. You understand the fundamentals of linear algebra, stochastic gradient descent, and can derive linear regression by hand.
Post-training & Environments. You are comfortable running SFT across distributed GPU clusters using FSDP. You have an appetite to improve existing models using high-quality data and are capable of reasoning about new environments and online RL (GRPO, PPO).
Python (FastAPI, PyTorch, etc.): REST APIs, back-end, vector-math. You can derive some high level abstractions in Torch using lower level primitives.
Experience: we don’t believe in hard and fast rules here. Roughly 3-5 years of experience is what we’re looking for, but more or less experience works as long as the above expectations resonate with you.
Automate your job search with Sonara.
Submit 10x as many applications with less effort than one manual application.
