Home › Jobs › Wizard Studios Inc
AI/ML Engineer
Wizard Studios Inc · San Francisco, California · Posted November 9, 2026 (10 days ago)
- San Francisco, California
- Full time
Job description
Ai/ml engineer
Location: San Francisco · In person
Wizard is the AI-native platform for creative work, starting with professional video: a native C++ editor and headless media runtime that humans, agents, and developers all operate.
We’re looking for a strong AI/ML generalist with hands-on experience building agent harnesses, designing agent evaluations, and working with multimodal models. You should be comfortable moving between research, experimentation, and production engineering, with a working understanding of post-training.
Why this role matters
Our go-to-market plan hinges on a five-minute demo: raw footage in, Oz assembles a cut, and a finished trailer comes out. The trailer’s quality is the test of whether the system works.
Making that happen requires reliable agents, useful representations of video and audio, and evaluations that distinguish a successful tool call from a successful creative outcome. You’ll help build those capabilities and turn research ideas into features that editors can use in real projects.
About the role
You’ll work across Oz’s agent harness, evaluation infrastructure, and multimodal understanding. That includes how the agent uses tools, manages context, interprets footage, recovers from failures, and evaluates the results of its edits.
This is a generalist role with substantial ownership. We’re looking for someone who can investigate a model limitation, design an experiment, implement a solution, and follow it through to production.
What you’ll do
- Build and improve agent harnesses. Develop tool execution, context management, task orchestration, recovery, and observability for agents working on complex editing tasks.
- Own meaningful agent evaluations. Build datasets and evaluation workflows that measure task completion, editing quality, reliability, latency, and cost. Use traces and failure analysis to identify regressions and guide improvements.
- Develop multimodal understanding. Work with video, image, audio, and text encoders to support retrieval, temporal understanding, and editorial decisions.
- Explore representation and fusion approaches. Evaluate how information from different modalities should be aligned and combined, including early and late fusion.
- Integrate models into the product. Choose and deploy local or API-based models according to quality, latency, memory, and operational requirements.
- Contribute to post-training experiments. Help identify useful training data and evaluate whether fine-tuning or other post-training approaches address recurring failures.
- Work closely with editors and engineers. Translate feedback from real editing sessions into measurable improvements.
Required qualifications
Must-have experience
- Strong practical experience building agent harnesses, including tool use, state and context management, and failure recovery.
- Experience designing agent evaluations that assess both execution behavior and end-to-end outcomes.
- A strong multimodal ML background, with familiarity across image, video, audio, and text encoders.
- Understanding of multimodal alignment, temporal representations, and early/late fusion techniques.
- Strong Python engineering skills and experience with a modern ML framework such as PyTorch.
- A working understanding of post-training, including training-data quality, fine-tuning, and evaluation.
- The ability to move between research prototypes and maintainable production systems.
Preferred qualifications
- Experience with video understanding, ASR, speaker identification, or audiovisual synchronization.
- Local inference optimization, model quantization, or distillation.
- Experience integrating ML systems with native applications or C++ runtimes.
- Experience evaluating creative outputs using automated measures and human judgment.
- Familiarity with professional video editing workflows.
Show us what you can do
We’d love to see systems you’ve built, research you’ve translated into working software, or evaluations that changed an engineering decision. Share code, papers, demos, or a concise account of a difficult failure you investigated and resolved.
About us
Wizard is the AI-native platform for creative work, starting with professional video. In roughly six months, our team built a native C++ editor and a headless media runtime that humans, agents, and developers all operate.
Our philosophy is creative as code: creative decisions should remain editable, controllable, composable, and reversible. We prioritize artist control and editorial quality.
Our sister company, Story, produces video for frontier AI labs and deep-tech companies. Story is our first customer and in-house design partner, and its editors cut real client work on Wizard every day.
Our team comes from Industrial Light & Magic, Netflix, Apple, Epic Games, Lucasfilm, Blizzard, and 2K. We work in person in San Francisco.
Compensation
The base salary range for this role is $175,000–$230,000. Your final offer will depend on your experience, skills, and level.
Posting bd8c4344-c0b4-4407-b67b-dd8f5b8eba43 on ashby. Source: the employer’s listing. Valid through December 24, 2026 unless closed earlier.