About this role.
This senior applied research engineering role develops production-grade foundation models for human-centric video generation. The work combines diffusion-model research, distributed multi-node training, rigorous evaluation, and inference optimization using Python, PyTorch, CUDA, DeepSpeed, AWS, and SLURM. The successful candidate will independently run high-signal experiments, improve model controllability and stability, and translate research outcomes into deployed product capabilities. It is a high-ownership role in a fast-moving Generative AI environment with strong expectations for reproducibility, scientific communication, and shipping impact.
Role DNA
A quick view of the complexity, pace, ownership and collaboration implied by the job description.
Job Complexity
5/5Pace & Pressure
5/5Autonomy Level
5/5Communication Load
4/5Salary analysis
Estimated compensation compared with the broader US market for similar roles.
Core skills
Skills and capabilities most closely associated with this opportunity.
Cover letter sample
Dear Synthesia Hiring Team,
I am excited to apply for the Senior Applied Research Engineer - Video Team role. My background in deep learning, PyTorch, distributed multi-GPU training, and experimental design aligns closely with your goal of building reliable production-grade video foundation models.
I am particularly motivated by the opportunity to improve controllability, training stability, evaluation quality, and inference efficiency for human-centric generative video systems. I bring an outcome-focused approach to research: forming clear hypotheses, running reproducible ablations, interpreting noisy results, and translating validated advances into scalable product capabilities.
I would welcome the opportunity to contribute to Synthesia's high-ownership team and help ship impactful generative AI technology.
Sample interview questions
I would outline the model architecture, dataset, compute scale, and measurable goal, then explain how I diagnosed issues such as instability, memory pressure, or low throughput. A strong answer should include concrete interventions—for example, mixed precision, gradient checkpointing, FSDP sharding, optimizer tuning, or data-pipeline improvements—and quantify the resulting quality, stability, or efficiency gains.
I would define a baseline and change only the conditioning mechanism while holding data, training budget, seeds, and evaluation protocol constant. I would measure automated fidelity and temporal-consistency metrics alongside structured human evaluations focused on instruction adherence, pose accuracy, and emotional expression, then use confidence intervals and failure-case analysis to determine whether the improvement is meaningful.
I start by verifying reproducibility and isolating whether the issue is numerical, data-related, systems-related, or caused by the optimization setup. I inspect loss and gradient statistics by rank, validate synchronization and effective batch size, check mixed-precision overflow behavior, and reduce the system to a smaller reproducible configuration before reintroducing scale. I also use profiling and monitoring to identify communication bottlenecks, uneven data loading, or memory fragmentation.
I would first profile end-to-end latency to separate model, decoding, data-transfer, and serving overhead. Depending on the bottleneck, I would evaluate fewer-step or distilled samplers, compilation and kernel optimization, batching, quantization where quality permits, caching, and model architecture changes such as more efficient attention. Every optimization would be validated against quality, controllability, temporal consistency, throughput, and cost targets.
A strong example would show that I established success criteria before investing heavily, ran a targeted experiment, and recognized that the result did not justify further compute or engineering time. I would communicate the negative result clearly with the hypothesis, methodology, evidence, limitations, and recommended next step, so the team could preserve the learning and redirect effort quickly.
Synthesia is the world’s leading AI video platform for business, used by over 90% of the Fortune 100. Founded in 2017, the company is headquartered in London, with offices and teams across Europe and the US.
As AI continues to shape the way we live and work, Synthesia develops products to enhance visual communication and enterprise skill development, helping people work better and stay at the center of successful organizations.
Following our recent Series E funding round, where we raised $200 million, our valuation stands at $4 billion. Our total funding exceeds $530 million from premier investors including Accel, NVentures (Nvidia’s VC arm), Kleiner Perkins, GV, and Evantic Capital, alongside the founders and operators of Stripe, Datadog, Miro, and Webflow.
About the role
As a Research Engineer in our Video team, you will help build the next generation of production-grade foundation models for human-centric video generation.
You will join a highly focused team working at the intersection of large-scale generative modeling, distributed systems, and production engineering. Our mission is to develop and optimize video base models that power realistic, controllable, and emotionally expressive synthetic humans at scale.
This is not pure research. This is applied research with direct product impact.
You will work on advancing training recipes, scaling distributed systems, improving evaluation frameworks, and optimizing inference to ensure our models are high quality, stable, and efficient enough for real-world deployment. Your work will directly influence models used by tens of thousands of businesses worldwide.
What you’ll do
You will own and execute end-to-end research and engineering projects, from hypothesis to production impact. This includes:
Developing and scaling latent video diffusion models tailored for human-centric video generation
Designing conditioning mechanisms to improve control (pose, emotion, script, camera) without sacrificing fidelity
Advancing distributed training strategies (DDP, FSDP, DeepSpeed, sequence parallelism) under real compute constraints
Improving training stability at multi-node scale
Designing rigorous evaluation frameworks combining automated metrics and structured human evaluation
Optimizing inference for low latency, high resolution, and cost efficiency
Running controlled ablations and experiments to drive high-signal modeling decisions
Contributing to high engineering standards: reproducibility, experiment tracking, CI/CD, monitoring
You will be expected to move fast, run multiple hypotheses in parallel, identify signal early, and focus on outcomes rather than exploration for its own sake.
What we’re looking for
Must-have
Strong experience training deep learning models at scale
Strong Python and PyTorch skills
Hands-on experience with diffusion models (image domain required; video preferred)
Experience with large scale multi-GPU / multi-node training
Good understanding of distributed training (DDP, FSDP, DeepSpeed or similar)
Ability to design controlled experiments and interpret noisy results
Nice-to-have
Experience with video diffusion models
Experience in avatar or human-centric generation
Familiarity with world / interactive models
Experience with GANs or VAEs
Experience optimizing inference systems for production
Our stack
Python, PyTorch, CUDA
DeepSpeed, distributed training & inference
Sequence parallelism
AWS, SLURM, Docker
GitHub, CI/CD pipelines
Who you are
You are research-driven but outcome-focused
You care about shipping, not just publishing
You can explore multiple ideas quickly and drop low-signal directions early
You communicate clearly and present results scientifically
You operate independently but collaborate actively across teams
Why join us?
Build production-scale video foundation models in a fast-growing Generative AI company
Work on human-centric video generation with real-world impact
Tackle hard problems in scaling, stability, and controllability
Influence the direction of next-generation synthetic human technology
Join a highly technical, high-ownership environment where your work ships
If you want to work on cutting-edge generative video models and see your research power real-world products, we’d love to talk.
Our culture
At Synthesia we’re passionate about building, not talking, planning or politicising. We strive to hire the smartest, kindest and most unrelenting people and let them do their best work without distractions. Our work principles serve as our charter for how we make decisions, give feedback and structure our work to empower everyone to go as fast as possible. You can find out more about these principles here.
Serving 50,000+ customers (and 50% of the Fortune 500)
We’re trusted by leading brands such as Heineken, Zoom, Xerox, McDonald’s and more. Read stories from happy customers and what 1,200+ people say on G2.
Proprietary AI technology
Since 2017, we’ve been pioneering advancements in Generative AI. Our AI technology is built in-house, by a team of world-class AI researchers and engineers. Learn more about our AI Research Lab and the team behind.
AI Safety, Ethics and Security
AI safety, ethics, and security are fundamental to our mission. While the full scope of Artificial Intelligence’s impact on our society is still unfolding, our position is clear: People first. Always. Learn more about our commitments to AI Ethics, Safety & Security.
The good stuff…
Competitive compensation (salary + stock options + bonus)
Fully remote from Europe or hybrid work setting with an office in London, Amsterdam, Zurich, Munich
25 days of annual leave + public holidays
Great company culture with the option to join regular planning and socials at our hubs
+ other benefits depending on your location
You can see more about Who we are and How we work here: https://www.synthesia.io/careers
Annual salary information is not provided for this position. Explore salary ranges for similar roles in our Salary Directory ›
Upload your resume now
- To unlock remote work opportunities and be discovered by global employers.
This job listing has been manually reviewed by the Jobicy Trust & Safety Team for compliance with our posting guidelines, including verification of the company's legitimacy, accuracy of job details, clarity of remote work policy, and absence of misleading or fraudulent content.
