Production Human-Motion Platform
Cinamon · AI Lead · 2023–2026 — Took motion generation, retrieval, refinement, and retargeting from data pipelines to optimized inference.
Product problem
Creative 3D products need more than a motion model. They need a connected platform for preparing skeletal data, generating and retrieving motion, improving transitions and locomotion, converting between rigs, and serving models reliably inside product workflows.
Role and ownership
As AI Lead at Cinamon, I worked hands-on across motion-model research, data and training infrastructure, and production inference while setting priorities around product requirements and research opportunities.
What I shipped
- Developed text-based motion generation and retrieval systems, including refinement and selection paths for product use.
- Built locomotion, motion-transition, and cross-rig retargeting capabilities for controllable character animation.
- Connected motion-data preparation and validation to reusable model-training workflows.
- Exported and served motion models through Triton and ONNX-based inference paths.
Engineering depth
Motion-text data collection, skeletal-motion preprocessing, motion representations, SMPL and production-rig conversion, multi-GPU training, model export, batched inference, and service integration.
Public outcome and evidence
The resulting platform covered the lifecycle from motion-data acquisition and model development to optimized inference for creative 3D workflows. COMET provides public evidence of how a product motion-control problem was developed into released research and code.