Research Scientist – Interactive Avatars

Overview

As a Research Scientist in Synthesia’s Interactive Avatars team, you advance diffusion-based, avatar-centric video models that listen and respond in real time. You will shape dyadic interaction modeling and push the perceptual and multimodal capabilities of interactive agents from research to product. You’ll work in a fast, high-ownership environment to deliver robust, production-scale capabilities that power human-centric video generation.

Responsibilities

• Define and drive research direction for dyadic interaction modeling end-to-end
• Advance perceptual understanding of user audio/video and generate appropriate reactions
• Post-train multimodal models to create natural dyadic interactions from audio/video inputs
• Adapt diffusion models to conditioning signals like conversational state and listener cues
• Build evaluation frameworks and test suites for continuous interaction quality tracking
• Collaborate with data team to define data needs and curate high-quality datasets
• Conduct rigorous experiments and communicate findings to influence technical decisions

Key requirements

• Strong ML background with hands-on diffusion models experience for video or avatar generation
• Publications in top-tier venues or demonstrated impact in world models, dyadic interaction, or video diffusion
• Experience converting research ideas into working implementations and production readiness
• Proficiency in PyTorch and modern ML tooling for large-scale training
• Clear communication of hypotheses, experiments, and results

• Clear communication of complex ideas
• Collaborative and cross-functional teamwork
• Ability to run and document rigorous experiments
• Diffusion models (video/avatar generation)
• Multimodal modeling
• Conversations modeling (turn-taking, listener cues)

…