I lead agentic development and AI for NVIDIA Developer Programs.

I started in robotics and computer vision, getting machines to see and act. That grew into GPU video pipelines, physical AI and vision-language models on Jetson and DGX. Along the way I helped about 70 startups ship vision AI on NVIDIA hardware and taught NVIDIA DLI courses. Now I build agents, the frameworks under them and the inference behind them. Most of that code is internal to NVIDIA, so this page is the summary.

I love developer tools, local AI and open source.

Now

• Coding agents: Claude Code, Codex, Cursor and Devin, usually all at once, orchestrated from T3 Code. Each gets its own worktree and the right model for the job, and nothing merges until I approve it. The skills, hooks and plugins they run on are mine.
• Building agents: I built and run the platform behind NVIDIA Developer Programs' Slack agents: typed Pydantic AI agents with evals, prompts tuned with DSPy and GEPA, and tools over MCP.
• Physical AI: DeepStream pipelines feeding vision-language models (Nemotron, Cosmos) and multi-view 3D tracking, measured on Jetson, DGX Spark and discrete GPUs.
• Inference: vLLM, SGLang and TensorRT-LLM, tuned for time to first token.
• Local AI: my house has its own agent, built on the open-source Hermes Agent. It runs on a Jetson AGX Orin next to Home Assistant, Scrypted, Whisper and Kokoro, and thinks with an open-weight Qwen model served from a GB10 box on the LAN.

Stack

$ guy --stack
devtools    t3-code  claude-code  codex  cursor  devin
agents      pydantic-ai  dspy  gepa  mcp  litellm
vision      deepstream  tensorrt  triton  jetson  cupy  opencv
inference   vllm  sglang  tensorrt-llm  nim  llama.cpp
homelab     hermes-agent  home-assistant  scrypted  whisper  kokoro  qwen
core        python  rust  c++  cuda  pytorch
infra       aws  terraform  docker  kubernetes  gitlab-ci

Fun facts

• Agents write most of my code. I review all of it.
• I'd rather delete a part than add one.
• I lift heavy things. 🏋🏻

guyda.dev · linkedin · x