The company is building the base layer for visual intelligence: multimodal models trained together on images, video, audio and action so they can see, generate and understand the world.
The founding team pioneered latent diffusion, the method behind generative AI tools that reached millions of people.
Current work covers generative media at scale, machines that act in the physical world, and computer use.
The models rank at the top of public benchmarks and run in production on some of the largest platforms globally.
The company is a small, quickly growing research lab competing with the biggest companies. It has labs in Freiburg and San Francisco.
This role is a Forward Deployed Engineer position focused on ML infrastructure and inference.
Responsibilities
Create custom endpoints and pipelines on the company's models for customers, such as multi-step workflows, prompt rewriting and upsampling, pre- and post-processing, moderation and safety layers, and pairings with VLMs and other models.
Guide a customer's use case from initial scoping through a proof of concept to a production endpoint, on either the company's or the customer's infrastructure.
Handle the rollout of models and custom pipelines into customer production, from on-premise to company-hosted setups, including serving, GPU sizing and scaling.
Analyze customer workloads and apply current inference optimizations to meet cost and latency goals, working with the inference team on deeper optimization.
Build evaluation sets and harnesses based on customer needs, and detect quality drops between model versions.
Help customers tailor models through finetuning and LoRA training pipelines, run on their infrastructure or the company's.
Advise on deployment architecture, weighing company-hosted against self-hosted options for cost, latency, security and compliance.
Pass field findings to research and platform teams, and turn repeated customer work into shared tools, reference architectures and documentation.
Serve as technical lead within the FDE team, set standards for applied and deployment work, and mentor other engineers.
Requirements
6+ years of engineering experience, including substantial time building and running ML systems in production.
Hands-on work building and serving models in production, from model code to API, using PyTorch, serving frameworks and GPU workloads.
Experience building end-to-end applied ML systems that chain several models together with evaluation and safety or moderation parts.
Solid infrastructure skills: Kubernetes, containers, observability, and at least one major cloud (AWS, GCP or Azure).
Experience working with customer engineering teams in complex enterprises, including security reviews and work in regulated or air-gapped environments.
Strong Python skills and comfort reading and changing model code.
Can explain complex technical trade-offs clearly to engineers and non-technical people, both inside and outside the company.
Willing to act as Technical Deployment Lead (TDL) for strategic accounts, owning the technical relationship and delivery end to end.
Willing to travel occasionally within the country and internationally for team offsites and customer work.
Prior knowledge of diffusion models and/or flow matching, plus finetuning and distillation techniques.
Inference optimization experience, with a record of improving latency, throughput or cost per generation.
Experience using LLMs or VLMs for prompt harnesses, captioning or evaluation.
Experience with open source tooling from the image model ecosystem, such as ComfyUI, diffusers and common trainer frameworks.
Experience serving video generation models.
Open-source contributions to inference, serving or diffusion model projects.
Previous work as a forward deployed engineer, solutions architect or ML platform engineer at an AI lab or cloud provider.
Conditions
Based in San Francisco, with in-office presence at least 2 days a week, or one full week every other week.
Salary is USD 273-395K, depending on profile.
The team works around four values: high standards for science and product, low ego with shared credit, bold shipping over waiting for perfect conditions, and kindness with empathy.