Machine Learning Engineer

We're building an end-to-end platform for reinforcement fine-tuning. We help the fastest growing AI companies fine-tune OSS models that outperform foundation models.

We bring a combination of deep startup and technical expertise:

Kasey (CEO) - Previously co-founded and sold a gaming startup, most recently worked in early-stage VC focusing on AI investments. In a past life, he competed in classical piano competitions and performed at Carnegie Hall.

Andy (CTO) - Real-time data/ML expert who was the youngest tech lead at TikTok, where he led their real-time recommendations & data infrastructure team. Hacked Gradescope during COVID.

We've raised $7M from leading institutional investors like YC & CRV, as well as angels like Paul Graham and Guillermo Rauch.

At Osmosis, we help companies use cutting-edge reinforcement learning techniques to fine-tune open-source language models that beat foundation models on performance, latency, and cost.

We’ve raised $7M in funding from Y Combinator, top institutional investors like CRV and Audacious Ventures, as well as angel investors including Paul Graham (Y Combinator), Erik Bernhardsson (Modal Labs), Misha Laskin (Reflection AI), and Guillermo Rauch (Vercel).

We're looking for a Machine Learning Engineer to contribute to high-performance distributed training infrastructure for RL at scale. You'll work directly with our founding team and design partners to push the boundaries of what's possible with post-training and continual learning systems.

This role requires expertise in RL algorithms, distributed training, and low-level optimization. You'll have exceptional agency to make impactful decisions while working in a fast-paced, customer-driven environment.

You’ll contribute to work in areas like:

• Distributed Training Infrastructure: implement new RL algorithms and build scalable post-training pipelines
• Resource Management & Optimization: design infrastructure systems for efficient GPU utilization and dynamic resource allocation
• Customer-Facing Work: work directly with customers on production deployments and custom model development

• Backend: Python FastAPI, Golang
• Frontend: React, TypeScript, Next.js
• Cloud Infrastructure: AWS Fargate, Docker, Kubernetes, AWS SageMaker
• ML Frameworks: Verl / slime / Megatron-LM / SkyRL, PyTorch (FSDP experience is a plus), vLLM / SGLang
• Databases: DynamoDB, S3

Backend: Python FastAPI, Golang Frontend: React, TypeScript, Next.js Cloud infrastructure: AWS Fargate, Docker, Kubernetes, AWS SageMaker ML frameworks: Verl (nice to have), PyTorch Databases: DynamoDB, S3, LanceDB

1. 30 minute chat with Kasey (CEO)
2. 30 minute chat with one of our MLEs
3. 60 minute technical screen with Andy (CTO)
4. 1-3 day paid in-person work trial
5. References and final decision!