About the job
We are seeking a Machine Learning Scientist to lead the research and development of Large Language Models (LLMs), Vision-Language Models (VLMs), and multi-modal foundations and solutions for games. This role is defined by a mandate for inference efficiency; you will not only build and fine-tune state-of-the-art models but also lead the algorithmic innovation required to make them viable in terms of cost, latency, and quality across a variety of cloud and edge devices.
Responsibilities
Model Adaptation & Alignment: Design and own the fine-tuning and alignment of LLMs and VLMs in PyTorch, leveraging modern preference learning and reinforcement learning to enhance reasoning, tool-use, and agentic workflows for interactive game systems.
Algorithmic Model Optimization: Lead efforts in model compression—specifically knowledge distillation, structural pruning, and architectural refinement—to create efficient variants of large models that meet strict latency, cost, and quality constraints.
Generative Visuals & Diffusion: Develop and optimize Diffusion-based models for Image, Video, and 3D generation, including distillation and efficiency techniques for viable game-time performance.
Pragmatic Model Integration: Strategically evaluate and integrate SOTA open-source and commercial models while building internal "layers," adapters, and enhancements to fill gaps in creative control.
Multi-modal Interaction: Optimize and integrate audio (ASR/TTS), language, and vision models to enable low-latency, cross-modal reasoning and interaction.
Qualifications
Minimum
Multi-modal Architecture Expertise: Strong foundation in deep learning architectures, with deep expertise in Transformers and Diffusion architectures powering LLMs, VLMs, and generative visuals, including their specific performance bottlenecks.
Optimization Specialist: Proven track record in algorithmic model optimization (e.g., distillation, quantization-aware training, or pruning) to reduce FLOPs and memory footprint.
Data-Centric Mindset: Skilled in data cleaning, curation, and the creation of synthetic data for complex evaluation and training pipelines.
Pragmatic Builder: Ability to prioritize impact by deciding when to use commercial APIs/OSS weights versus when to invest in proprietary R&D to solve efficiency or quality problems.
Programming: Expert proficiency in Python and deep learning frameworks (such as PyTorch); ability to collaborate with engineering on low-level performance constraints.
Preferred
Prior experience optimizing models for heterogeneous hardware (Mobile, Cloud GPU, and custom edge devices).
Expertise in audio-visual multimodal models and video generation.