Staff Research Engineer - Multimodal Generative Modelling
Synthesia · Remote · Europe ·
COMPENSATION
€•• – ••k
Listed on the employer's job board (Ashby)
- Staff level role
- Fully remote options
- PyTorch + LLMs
- Multimodal AI
- London HQ
Join Synthesia's R&D department to define and drive the vision for human-interactive models. You will design systems that perceive and respond to user actions and emotions, moving beyond simple turn-taking to create natural, real-time conversations.
Working with the Voice team, you will propose research directions and own the implementation of critical components. Your work will integrate text and voice models to enhance visual communication for over 60,000 businesses.
What you'll do
- Shape the roadmap for new model capabilities and customer functionality
- Propose novel multi-modal system architectures for text and voice
- Develop low-latency streaming systems for interactive voice-video synthesis
- Design solutions to improve emotional expressiveness and natural interaction
- Implement designs from pretraining through post-training phases
- Integrate neural codecs, diffusion, and flow-matching architectures
- Define new evaluation metrics for latency and interaction quality
- Curate new datasets to complement existing training data
- Lead post-training initiatives including DPO, fine-tuning, and distillation
- Ship optimized models to production and address customer feedback
What you'll need
- Ability to advance interactive multimodal systems with novel ideas
- Strong understanding of generative modelling for sequential or multimodal data
- Hands-on experience with large language models or transformer architectures
- High proficiency in PyTorch, distributed training, and model optimization
- Solid grasp of time-series modeling and tokenization for audio or video
- Proven ability to prototype quickly and iterate on hypotheses efficiently
- Experience training deep learning models end-to-end from data to evaluation
- Strong software engineering skills for shared research infrastructure
Nice to have
- Shipped a generative model into a live product at meaningful scale
- Experience with conversational systems where latency is a primary constraint
- Worked on LLMs with large-scale training and reasoning capabilities
- Owned a research problem from architecture to production deployment
- Collaborated across audio, video, research, and product teams
- Experience with real-time or streaming architectures
- Familiarity with diffusion, neural codecs, or autoregressive decoders
- Original research contributions published at top-tier venues
Stack
- PyTorch
- Large Language Models
- Transformers
- Neural Codecs
- Diffusion Models
- Flow-matching Models
- Autoregressive Decoders
How you'll work
- Headquartered in London with teams in Europe and the US
- Part of a 40+ person R&D department
- Collaborate closely with Voice lead and video teams