~/greenbaumlabs $ status --all

Greenbaum
LabsA personal AI workshop running on local compute.

Side projects, diagnostic tools, and experiments in applied AI—built on my own machines, wired together like a small internal platform. Three boxes on a 10 GbE backbone. Compute stays local, the filesystem is the source of truth, and dashboards are visibility layers, not authority.

For operators, engineering leaders, and Chiefs of Staff who care less about AI demos and more about reliable, explainable systems that actually help people do better work.

If something here feels adjacent to a problem you're wrestling with, I'd welcome a conversation: justin@greenbaumlabs.com

Scroll
Infrastructure

The Workbench

Greenbaum Labs is the workshop. These are the other rooms: three machines doing the quiet, unglamorous work of running local models, agents, and diagnostics.

Compute
DGX Spark
Main box: primary dev environment, orchestration, and most of the day-to-day experiments.
NVIDIA Grace Blackwell
128GB unified memory
Local inference · Fine-tuning
Storage
Synology NAS
Storage / backbone: snapshots, logs, and the slow, boring parts that make the rest trustworthy.
Filesystem-first truth store
10 GbE backbone
Markdown canon · Schemas
Interface
Mac Studio
Experiment box: throwaway agents, risky ideas, and things that might crash.
Primary workstation
10 GbE backbone
Claude · Cursor · Terminal
┌─────────────────────────────────────────────────────────┐ │ GREENBAUM LABS · Network Topology │ ├─────────────────────────────────────────────────────────┤ │ │ │ ┌──────────┐ ┌──────────┐ │ │ │ DGX Spark│─10GbE──│ NAS │ │ │ │spark-a044│ r/o │ gl_nas │ │ │ └────┬─────┘ └────┬─────┘ │ │ │ │ │ │ │ ┌──────────┐ │ │ │ └────│USW Flex │───┘ │ │ │ XG │ │ │ └────┬─────┘ │ │ │ │ │ ┌────┴─────┐ │ │ │Mac Studio│ │ │ │ control │ │ │ └──────────┘ │ │ │ └─────────────────────────────────────────────────────────┘
Projects

What's on the bench.

Things I'm building, things I've shipped, and things I'm still thinking through. Not all of them are finished—that's the point of a workshop. Each project exists because of a real tension I've seen in organizations: invisible risk, narrative drift, siloed data, or teams flying blind. The lab is where I prototype ways to see and fix those problems earlier.

GL-001
Coherence Diagnostic Engine
Leaders sense something is off but can't name it. Patterns repeat across teams without shared language.
Semantic search across years of decision logs and observations. Surfaces recurring tensions without deciding what they mean.
DGX Spark ChromaDB Local LLM Decision Support
Live
Case Study →
GL-002
RC Pit Boss
RC race directors lose track of timing, heats, and standings during chaotic race days.
Voice-controlled race management that handles marshals, timing, and announcements. Built for one person, shipped as a gift.
Python Whisper Voice UI
Live
GL-003
Brand Ecosystem Builder
Personal brand assets scattered across platforms with inconsistent voice and visual language.
Unified design system across justingreenbaum.com, Substack, photography portfolio, and Labs. Single source of truth for brand.
Claude Design System
Live
justingreenbaum.com
GL-004
50K Image Index
A decade of photography sitting invisible on drives. No way to find what I shot or when.
AI-assisted cataloging with metadata extraction, location tagging, and semantic search across the full archive.
DGX Spark CLIP Metadata
Planned
GL-005
Pattern Surfacer
Important signals buried in daily logs. Patterns only visible in retrospect.
Reads logs and flags recurring signals without interpretation. "These three things appeared together four times this month."
DGX Spark Local LLM NAS
Planned
Operating Principles

How the lab works.

01
Own the compute. Local inference on local hardware. The DGX isn't a luxury, it's a design constraint. When you control the infrastructure, you control the dependencies.
02
Filesystem is source of truth. Dashboards are visibility layers, not authority. The NAS holds the canon. Everything else reads from it.
03
Automation may observe, summarize, and suggest. Automation may not decide. This is AR-001. Every tool in this lab operates under this constraint.
04
Ship gifts, not pitches. RC Pit Boss was built for one person. The best projects start as something useful for someone specific.
05
Build ahead of demand. The work will become important. The question is whether the infrastructure is ready when the need arrives.
The Ecosystem

Where this fits.

Greenbaum Labs is the workshop. These are the other rooms.

The Thesis
Justin Greenbaum
Decision and responsibility infrastructure. Advisory work, AI governance, and the Coherence framework.
justingreenbaum.com
The Writing
Substack
Essays on coherence, decisions, and building infrastructure for what matters.
writing.justingreenbaum.com
The Eye
Greenbaum Photography
Documentary and fine art photography on Leica. Five continents, 50K+ images.
greenbaumphotography.com
The Studio
Megabrixels
Fine art photography meets miniature worlds built from Lego. Craft as antidote to content.
megabrixels.com
~/connect

Open garage door.

I've spent most of my career close to operations, risk, and workflows. Greenbaum Labs is where I test what happens when you give that kind of work a small, stubborn AI workshop instead of another slide deck.

If you're exploring an internal AI lab, advisor role, or just want to compare notes on how to wire these systems together on real infrastructure, send a note my way.

justin@greenbaumlabs.com
LinkedIn Substack GitHub