Research Engineer -- Vision and Vision-Language Models

Zyphra; San Francisco (CA); On-site; USD 180,000 - 240,000; Full time

Benefits offered by this job

Comprehensive medical, dental, vision

Competitive compensation

Relocation and immigration support

In-office snacks and meals provided

Unlimited PTO and holidays

In-person team in San Francisco

Job summary

Zyphra is seeking a Research Engineer - Vision & Vision-Language Models to join the Vision Team in San Francisco. You will work across the full model development stack, from data sourcing to pretraining, post-training, RL, and evaluation.

Experience with vision-language models is valuable but not required, and you will be evaluated on learning speed and execution. You will contribute to end-to-end training, model development, and dataset creation, collaborating with researchers and engineers

Qualifications

• Experience with vision-language or multimodal models is valuable.
• Strong ability to prototype ideas into experiments quickly.
• Excellent communication and collaboration in a fast-paced research setting.
• Experience with large-scale pretraining, post-training, or RL is a plus.

Responsibilities

• Lead end-to-end pretraining and post-training of vision-language models.
• Run large-scale vision and multimodal training experiments.
• Source, generate, curate, and evaluate vision datasets.
• Explore RL and other post-training methods for vision-language models.
• Contribute to architecture, data, and training methodology improvements.
• Collaborate across research and engineering to scale implementations.

Skills

PyTorch

Python

Multimodal models

Research prototyping

Experimentation

Communication

GPU clusters

Education

Postgraduate study in CS/ML

Tools

Web frameworks

Job description

The Role:

As a Research Engineer - Vision & Vision-Language Models, you will be a core contributor on Zyphra’s Vision Team, building the next generation of models for image understanding, computer use, and other agentic visual tasks.
You will work across the full model development stack, from sourcing and generating data to pretraining, post-training, reinforcement learning, and evaluation. The Vision Team owns its models end-to-end, and you will have the opportunity to contribute wherever you can have the greatest impact.
We care strongly about rate of learning, research taste, and ability to execute. Previous experience training vision-language models is valuable, but not required.

You’ll Work Across:

• End-to-end pretraining and post-training of vision-language models
• Large-scale vision and multimodal training runs
• Vision and multimodal dataset sourcing, generation, curation, processing, and evaluation
• Reinforcement learning and other post-training methodologies for vision-language models
• Computer use, visual agents, and other multi-turn agentic vision tasks
• Architecture, data, and training methodology ablations and improvements
• Performance optimization of our training and data processing stacks

What We're Looking For / Requirements:

• Strong research taste and intuition, with the ability to take ambiguous problems from conception to experimentation and execution
• Strong implementation and prototyping ability; you can turn ideas into experiments quickly
• A high rate of learning and an ability to rapidly become effective in unfamiliar technical areas
• Strong general machine learning intuition and an interest in understanding the full model development stack
• The ability to work well with others in a high-paced research setting
• Excellent communication and collaboration skills, and the ability to work effectively across both research and engineering implementation at scale

Qualifications / Additional Skills:

• Experience working with vision, vision-language, video, or other multimodal models or datasets is highly valuable
• Experience training vision-language models is a plus, but not required
• Experience with large-scale model pretraining, post-training, reinforcement learning, or related methodologies
• Experience sourcing, generating, or curating vision datasets, including synthetic data generation
• Familiarity with image processing tools, libraries, or pipelines
• Experience with computer use agents, visual agents, GUI understanding, or other multi-turn multimodal tasks is a plus
• Experience with frontend development, web frameworks, UI/UX implementation, or mobile development is valuable, particularly for computer use and interface-understanding tasks
• Experience training models on large-scale, multi-node GPU clusters
• Strong grasp of experimental methodology for running rigorous ablations and hypothesis testing
• Understanding of large-scale, highly parallel data processing pipelines
• Proficiency with PyTorch and Python
• Strong ability to dive into large pre-existing codebases and rapidly get up to speed
• Published machine learning research in well-respected venues is a plus
• Postgraduate study in a scientific or technical field such as Computer Science, EE/EECS, Mathematics, Physics, or Machine Learning is a plus, but we care more about demonstrated ability and trajectory than credentials

Why Work at Zyphra:

• Our research methodology is grounded in methodical, step-by-step approaches to ambitious goals. Both deep research and engineering excellence are equally valued
• We strongly value new and crazy ideas and are very willing to bet big on new ideas
• You will have the opportunity to work across the entire model development lifecycle rather than being constrained to a narrow part of the stack
• We move as quickly as we can; we aim to minimize the bar to impact as much as possible
• We value rate of learning and growth, not just what someone has already done
• We all enjoy what we do and love discussing AI

Benefits and Perks:

• Comprehensive medical, dental, vision, and FSA plans
• Competitive compensation and 401(k) plan
• Relocation and immigration support on a case-by-case basis
• In-office snacks and meals provided
• Unlimited PTO and company holidays
• In-person team in San Francisco with a collaborative, high-energy environment