Character & Simulation TA / Applied-AI Engineer
For 25 years I've rigged and simulated digital characters for feature film and AAA games, from The Matrix to World of Warcraft. Today I build applied-AI agents and the evaluation systems that prove they hold up.
Most of those years I led the work as well as doing it: the Avatar Hair team at Meta, seven years as Rigging & Simulation Lead on Blizzard cinematics, and a CharacterFX team of 25+ at DreamWorks. The job was always systems rather than shots, one rig or solver that has to hold up across an entire slate of characters.
Simulation teaches one lesson that transfers completely: a result can look right and still be wrong. A cloth sim that reads perfectly in one frame can be broken in every other. Agentic and generative systems fail the same way, plausibly. So I build agent systems and retrieval pipelines together with the evaluation harnesses that prove they still work.
A demo proves an agent worked once. A harness proves it is getting better. I build the harnesses. At Meta I hardened a 50+-user deep research swarm agent against a 15-question scoring matrix: 200+ full rounds of measure, fix, re-score. I also shipped devserver task loops that wrote their own self-improving skills, and team project frameworks putting engineers and OpenClaw in one shared cloud workspace.
An independently authored realtime runtime for Hermes Agent on LiveKit. Speech answers in milliseconds beside agent work that runs for minutes, each with its own cancellation scope, so interrupting a sentence never kills the task behind it. The current layer is task control: dispatch that fails closed without a real delegation handle, stop by exact handle, and request-bound command approvals. Experimental.
Architecture noteA UE 5.8 project where one skeleton-agnostic Animation Blueprint Template drives a roster of differently-proportioned rigs. The original layer is verification: every rig's graph is golden-diffed against a reviewed baseline and asserted per commit, with live per-node blend-weight introspection. Authoring composes the open-source monolith server.
Architecture notemainframe-mcp, a 100% local, GPU-accelerated knowledge server: Qwen3 embeddings with a native Qwen3 reranker, hybrid search, and ambient memory with conflict-aware consolidation. Its in-repo eval harness measured +26% / +33% retrieval gains.
Architecture noteclaude-oracle, a multi-tier research orchestrator. Parallel Haiku scouts fan out breadth-first, a Sonnet tier synthesizes, and one briefing comes back to the session: roughly 10x the research breadth at a fraction of the cost. MIT-licensed.
Architecture noteProblems single agents never hit. vram-mcp lets agents share one NVIDIA GPU: TTL claims, real busy detection via NVML, protected eviction. hardline-mcp gives local Claude Code, Codex and Hermes agents a durable mailbox.
Architecture noteA FLUX.2 LoRA that grows a full 94-character typeface from a two-glyph reference, vectorizes it into real Bézier outlines and ships an installable OTF and WOFF2. Candidates are picked by a four-metric harness, one of which I had to build.
Architecture noteSeven repositories are public on GitHub. Much of the rest lives behind NDAs: agentic pipelines and physics-ML training data at Meta, plus several private production systems. Glad to walk through any of it in depth.
Real-time avatar systems
Avatars & Horizon Meta Reality Labs · 2022–2026
Built the core avatar rigging and deformation systems in a production real-time VR/mobile engine: auto-rigging, autofit deformation, skinning, blendshapes and skeleton standardization, plus node-graph clothing-rig automation and runtime animation & simulation graphs, all inside hard VR and mobile performance budgets. Lead of the Avatar Hair team. Delivered 100 character rigs used to train a physics-based ML rigging model.
Rigging & Simulation Lead
Blizzard cinematics Blizzard Entertainment · 2015–2022
Full character rigging (full-body, facial, muscle, clothing and hair) across World of Warcraft, Overwatch, Diablo, Heroes of the Storm and Hearthstone. Architected the simulation branch of the studio’s modular, graph-based rig-build system and the PyQt tooling around it, and owned rigging and shot simulation on 25+ cinematics.
CharacterFX Lead
Eight feature animations DreamWorks Animation · 2005–2015
Led a CharacterFX team of 25+ while staying hands-on with hero-character cloth, hair and skin across How to Train Your Dragon, Kung Fu Panda, Puss in Boots, Bee Movie, Over the Hedge, Turbo, Penguins of Madagascar and Home. Pioneered cloth-based skin solvers for hero characters, and wrote many of the department’s core Python tools.
Autonomous crowd agents
The Chronicles of Narnia Rhythm & Hues Studios · 2004–2005
Motion editing for the autonomous crowd agents on The Lion, the Witch and the Wardrobe, plus motion-capture TD work.
Digital-human facial capture
The Matrix sequels ESC Entertainment · 2002–2004
Facial rigging, sculpting, match-move and texture projection on the Universal Capture system behind Reloaded and Revolutions, and part of the team behind that award-winning system. Primary contributor to “Making of The Superpunch”, the piece screened at the SIGGRAPH 2004 Electronic Theater.