Applied AI Engineer

About Kaitoi

Kaitoi is an AI-native orchestration platform for creative professionals. It is agentic across the entire surface (creative collaboration, execution, guidance, and live extension of the software itself) and self-evolving: users describe what they need and the platform shapes the workflow, builds the missing pieces, and grows around the shape of their work.

The platform is also built to be deeply production-friendly: a built-in local inference service, full MCP server and client, an HTTP API, and an automatic recipe system that turns model repos into portable, repeatable nodes. For someone building applied AI systems, the path from an idea to a capability that users actually run is short and direct.

The role

This is an applied AI role focused on making Kaitoi more capable, reliable, and useful for professional creative work. You will build and ship AI capabilities that improve real workflows: systems that understand creative intent, handle messy production inputs, run efficiently in practical environments, and turn advanced models into tools people can actually depend on.

We care about quality, but also practical deployment and usability. The work is not benchmark vanity; it is measured by whether creative professionals can get better results, faster, inside real production timelines.

You will own applied AI capabilities end to end: research direction, data strategy, prototyping, evaluation, inference optimization, integration, and shipping them as first-class parts of the platform.

What you will do

• Define and execute applied AI work that advances Kaitoi's creative capabilities
• Build and curate datasets that reflect real user workflows and production constraints
• Train, evaluate, adapt, and ship models in PyTorch where custom work is the right path
• Design honest evaluations against real creative material, not just synthetic metrics
• Optimize inference for the platform's local and hosted execution surfaces
• Wrap AI capabilities as Kaitoi nodes / recipes so they slot into existing user workflows
• Stay close to users and our in-house studio team. Their footage is the ground truth.

What you bring

• Strong PyTorch and deep familiarity with modern applied ML workflows
• A track record of shipping AI capabilities, not just papers or notebooks
• Practical instincts for data curation, evaluation design, and model iteration
• Comfort adapting research models into production systems
• Experience optimizing inference for real user environments
• Good taste around creative tools: you care about how people actually work, not just what a model can do in isolation

Nice to have

• Experience with image, video, audio, 3D, or multimodal creative systems
• Experience with local inference, model packaging, or deployment across varied hardware
• CUDA / Triton, ONNX, TensorRT, quantization, or other inference optimization experience
• Familiarity with professional creative pipelines and the constraints of production work
• Experience turning research code into maintainable product capabilities

How we work

• Fully remote, global. Async-first written communication.
• Small team, broad ownership, short feedback loops.
• We measure on real footage, ship to production frequently, and let users sit on top of what we build.

How to apply

Send a short note about an AI system or model you helped ship: what the problem was, how you built and evaluated it, and what surprised you. Links to code, papers, demos or before/after material go further than a full résumé.