I architect enterprise platforms for a living, and I run a distributed AI lab out of my house for the love of it. Every node, every GPU, every line of the stack: designed, built, and debugged by one set of hands.
SR
KNOWLEDGE 25 · LAS VEGAS
By day I am a ServiceNow platform architect: integrations, governance, platform strategy, and the unglamorous plumbing that keeps large organizations running. Fifteen years of SDLC discipline, from requirements to production.
The through-line in everything I build is the same: investigate before fixing, validate before shipping, and keep it lean. No bloat, no unnecessary dependencies, nothing that only works on someone else's computer.
Nights and weekends belong to Robinson Labs, where the same discipline gets applied to hardware I can actually touch.
Robinson Labs is a distributed AI platform running on consumer hardware: gaming GPUs, rescued server boards, and a rack's worth of machines I specced, assembled, cabled, and debugged myself. It serves 200 billion parameter models with a million tokens of context, entirely self-hosted.
The software stack on top is mine too: distributed inference across nodes, a model manufacturing pipeline, persistent memory for AI agents, local image generation, and the telemetry to keep all of it honest.
Frontier models earn a place in the toolchain too: every tool has a job. But the models this lab serves run on hardware I own, under my own roof.
Nothing here was bought as a solution. It was built as one.
Chat, tools, vision, and image generation, served from the lab's own GPUs. The front door to everything the cluster can do.
A pipeline for making models: pruning, fine-quantization, and modification, with provenance tracked on every artifact. Published models live on Hugging Face.
PostgreSQL-backed memory, decisions, and work tracking for AI agents, so every session picks up where the last one left off.
The lab's public face, including a daily AI news wire written and published by the lab's own agents. robinsonlabs.ai
Consulting inquiries, platform questions, or arguments about GPU pricing all welcome.