Senior Principal AI Software Engineer. I build agent systems that have to survive contact with real users: RAG platforms, local model serving, skills, and the hardware around them.

I run a local-first lab and open the pieces that are actually portable. I write in public as NeuralYogi.

• Production agents (Hermes, skills, MCP, local LLM routing)
• Hybrid RAG, evals, and internal platforms rather than chat demos
• Hardware you can print and flash: fail-closed FDM, LED matrix
• Local-first composition / wiki surfaces (house compiler stays private)

Current

• printables — Hermes FDM loop: intent → OpenSCAD or Blender → fail-closed DFM gates → P1S STL. Skill pack we actually run, not a slicer and not a dump of household parts.
• rage-matrix — Maker LEDDisplay2. 32×9 camera into sealed tank + war sims, title cards, HTMX hub.
• cyber-posture — portable Linux exposure scan plus host integrity / malware IoCs.
• agent-body-protocol — runtime-neutral body language for coding agents.
• recompose-experience — public UI prototype for a local-first composition layer. The house compiler and live hub are not in this repo.

Also useful

• ce-deep-agents — Compound Engineering workflows ported onto Deep Agents.
• deepagents-local — Deep Agents CLI on local Ollama.
• llm-benchmark — local LLM bake-off tooling.

I built and operated around Sigmanaut Mining Pool (dashboards, pool ops, MCP, reward tokens). That work is still public. It is not the main thread anymore.

• sigmanaut-mining-pool-ui
• ergo-mcp
• sigs-mega-core
• Mining-Reward-Tokens

Orgs from that era: @codeUTXO · @The-Last-Byte-Bar

Python · Docker · vLLM / llama.cpp · Hugging Face · RAG + MCP · OpenSCAD · C++ / ESP32

Public GitHub commits over time. Not a streak badge.

X @neuralyogi · neuralyogi.com · email

Tampa Bay, Florida.