Company: Sarama
Job: Senior ML Engineer
Type: Part time or Full time
Location: San Francsico, CA
Remote: Open to remote / Fully remote
Visa required: No
Required Experience: We want someone with experience in prompt engineering & fine-tuning models in PyTorch. Ultimately, this role will train, fine tune, and create a pipeline of a set of models that are ideal for a given project.
First Project: Given a video feed track a given subject and describe their actions. For example, given a basketball video of Lebron James create a fake ESPN highlight reel using a mix of custom models + API hosted models.
About Sarama: Sarama is a company that builds consumer apps using our infrastructure around developing low latency ML systems that leverage hosted models for optimal balance between power & speed. Our first app is a product that video records dogs and decodes their behavior + barks while learning the meaning based on input from users along with observed behavior. Our goal is to keep iterating on this app to ultimately decode all animal language from dolphins to dogs. The founders have both built the product with an early team of ML researchers that have had their work published in Nature, Science, & senior engineers that have built YC companies, scaled tech to be used by millions of people, and love writing software. If you like building with a strong group of balanced humans + dogs then Sarama is an amazing place. We have a company structure that balances heads-down all consuming periods with rest periods that is communicated ahead of time so everyone can plan their lives accordingly. We have 2 30-minute required meetings / week, everything else is scheduled as 1:1s or small group chats on Slack for project deep dives. We do a bunch of pair programming as well which we like.
About this role: We're looking for someone that's an incredible communicator vs a lone wolf that takes a project to finish in a vacuum. We enjoy working with people who love reading papers, learning new techniques, and continuously experiment/tinker with tech. We find that obsessive personalities do the best here and that someone who likes having an autonomous ability to build something, test it out, and iterate enjoy our culture. Lastly, we seem to attract people that tell us what we're doing is wrong (politely) and will find an approach that reduces complexity while improving user experience. Essentially our ideal candidate here has been tinkering with open source ML models especially around finetuning LoRA, training full models, and has an understanding of the trade offs required to get interesting results.
