What the role involves
Lucid is a small team building real-time interactive video models that spark joy, and are seeking a talented engineer to build the infrastructure that makes this possible. If you are passionate about our mission, we would love to hear from you!
Role Description
You are an experienced software engineer who builds the systems that power large-scale ML training. You understand how to move and process petabytes of data efficiently, manage distributed GPU workloads, and create infrastructure that researchers can rely on. You know how to optimize for the specific constraints of training large models while maintaining system reliability.
What you'll do
Build systems to handle petabytes of training data efficiently
Design and manage distributed training infrastructure across GPU clusters
Evaluate and implement orchestration systems (we currently use SLURM)
Work directly with ML researchers to identify and solve infrastructure bottlenecks
Build the foundational systems that let our team iterate quickly on large-scale experiments
Requirements
Strong distributed systems experience with focus on performance optimization
Deep understanding of ML training infrastructure and distributed training patterns
Experience with Python and PyTorch in production environments
Comfortable working with both cloud and bare metal infrastructure
Self-starter who can evaluate technical trade-offs and make architectural decisions
Excellent communication and collaboration skills
Location
San Francisco, CA
What we offer at Lucid
Interesting and challenging work
Base salary $180,000-250,000 plus equity
A lot of learning and growth opportunities
Health, dental, and vision insurance (US)
Regular team events and offsites
What they ask for
About Lucid
We are building universe simulations powered by interactive video models. We train video models that simulate hyper-realistic environments with immersive control, replacing hard-coded game or physics engines with dynamic neural networks. We built the fastest action-conditioned diffusion video model (running at 20+fps on a 4090 gaming gpu) to simulate minecraft. It is 5x faster than other minecraft World Models and was trained with 100x less resources. Our unique insight was relying on aggressive compression in our tokenizer (128x versus the traditional 8x), and because attention scales quadratically with # of tokens our model can run blindingly faster. Now we’re training a hyper realistic world model!
Full Lucid profile