Local-first AI inference
About 20,300 AI model calls ran on my own two Macs in October, served by a self-hosted 187 GB model behind a priority job queue I built and a library of nine open-weight models, with no per-call fees.
Lenovo Solution Sales Executive
Lenovo Solution Sales Executive based out of Chicago. Local-first AI enthusiast. Former college ballplayer, restaurant owner, avid golf trip attender, and the drawings are mine. Socials are down at the bottom if you want to connect.
Byrd Studio
Private things stay on my own hardware. The cloud still does the hard thinking — it just never sees the parts that are mine.
About
10 years in IT sales.
I grew up in restaurants, tending bar, playing baseball and drawing in class when I should have been paying attention. That's where the drawings on this page started. Today I'm part owner of the two restaurants below and an Airbnb.
My day job is sales. I've spent 10 years in IT sales at Lenovo, and I recently stepped into a new role as a Lenovo Solution Sales Executive, focused on professional services, software and AI. I'm based in Chicago.
At home I build local-first AI. Private work runs on my own machines, and the cloud only sees what it needs to.
- Sensitivity-aware model routing Every AI task gets a privacy label. In the last 30 days, local models answered 68 of 83 private tasks.
- Local-first AI inference A model I host myself finished 308 jobs in the last 30 days, with no per-call fees.
- MLOps and PII protection A small model that hides personal details retrains and is re-tested every night: 120 cycles so far, and 11 were rejected.
- Agentic AI, with guardrails Agents prep my meetings and draft follow-ups. They propose; plain code with an off switch decides what gets sent or saved.
Consulting
Freelance AI consulting. Free for now.
I take on a small number of AI projects on the side, with a network of engineers I trust for the parts I don't build myself. Local-first setups, agents that run unattended, and the guardrails that keep them honest.
It is free right now. I would rather do a few proofs of concept and earn the experience with you before anyone is charged. When it starts costing money, you will hear it from me first.
or message me on LinkedIn
or write to kabyrd012@outlook.com
Highlights
Local-first AI, running on my own hardware.
Agents and models I designed and operate myself, on two Macs and an edge device. Numbers counted on October 31, 2026.
Local AI agent fleet
18 AI jobs and services run entirely on my own hardware, including wiki ingest, nightly fine-tuning, voice-note review, inbox triage and a voice assistant. Seven more use local models as their partner or fallback.
Sensitivity-aware model routing
A routing layer labels each task by data sensitivity. Local models answered 319 private tasks, and no private task ever went to an outside backup provider.
MLOps and continuous fine-tuning
A nightly on-device train, test and promote pipeline has run 120 cycles. It only ships a model that leaks none of 467 private fields in synthetic test documents.
Edge AI and computer vision
A real-time person-detection model on a small edge AI device feeds 10 home automations, which fired 862 times in the last 30 days.
Context-aware automation
20 automations on a local hub read camera, room-sensor and phone-geofence signals to run 15 devices, with 1,151 actions in 30 days.
By the numbers
What runs at home, counted.
Counted fresh every time this page is rebuilt.
547automations and tools
1,692notes in my wiki
1,247ideas written down
148plans written
