Director, Multi-Modal AI: Agentic World Building

Unity Technologies•San Francisco, CA

About The Position

Generative AI has spent the last few years mastering pixels and prose. Unity is here to make it move. We are seeking a visionary Director, Multi-Modal AI to lead the charge in transforming "dream pixels"—static AI hallucinations—into Dream Games. This is a zero-to-one leadership opportunity to build Unity’s first-party intelligence layer. You won't just be calling APIs; you will be building the brain that understands 3D space, masters game mechanics, and empowers every creator to manifest their vision into a living, breathing Unity environment. You will lead the applied research organization responsible for Agentic 3D World Building. Your goal is to move beyond simple asset generation and solve the industry's "holy grail": Spatial Reasoning. By leveraging cutting-edge World Models and Multi-Modal LLMs, you will enable a future where natural language doesn't just describe a scene—it authors it, simulates it, and plays it.

Requirements

The "Spatial DNA": You don't just know AI; you know 3D. You understand that a game is a living system of transforms, raycasts, and rigidbodies.
Proven AI Leadership: 10+ years in ML/AI, with 5+ years leading high-performance applied research teams. You have a track record of shipping proprietary models that solve real-world problems.
Fine-Tuning & Distillation Experts: You are a master of instruction tuning, PEFT, and model compression. You know how to take a "generalist" model and turn it into a "Unity Specialist."
Agentic Visionaries: You have experience with agentic frameworks and autonomous systems that can reason through multi-step tasks in complex environments.
Optimization Obsessed: You understand the constraints of a real-time engine. You are comfortable with quantization (INT8/FP8) and speculative decoding to ensure the AI feels like a collaborator, not a bottleneck.

Responsibilities

Architect the "World Brain": Lead the research and implementation of Agentic workflows. You will use World Models to ensure AI understands the "why" and "how" of a 3D scene—from physics-based constraints to complex scene hierarchies.
Build the Multi-Modal Arm: Own the fine-tuning and distillation of state-of-the-art MLLMs. You will specialize these models to generate high-fidelity game mechanics, shaders, and logic that respond to multi-modal prompts (text, image, and spatial data).
Master the Teacher-Student Pipeline: Since we aren't just "pre-training" in a vacuum, you will drive the strategy for distilling massive foundation models into high-performance, edge-ready "students" optimized for the Unity Editor.
Bridge Science and Play: Solve the grounding problem. You will ensure that AI-generated content isn't just a visual trick, but is engine-ready, mathematically sound (transforms, quaternions, nav-meshes), and physically plausible.
Scale the Data Flywheel: Turn Unity’s massive ecosystem into an intelligence moat. Build secure, rights-managed pipelines and "teacher-led" synthetic data generation to train models on the nuances of game development that the rest of the world hasn't even seen yet.