AI first, human in charge. I treat AI as infrastructure: it does the tedious, error-prone work so people can do the judgment work. I read the symptoms, find the real problem and fix the cause, with zero AI slop. This page is the short version of what I built, what it runs on, and what I learned.
My 2024-to-date AI journey
I have been waiting my entire career for today. For decades I organized information, built systems and told stories with whatever tools I had. Since 2024 the tools have finally caught up with the way I think. A video archive close to 100 TB, with no reliable way to see what was inside it, became the reason to build better storage, a faster network, careful automation and local AI that reads, sorts and describes footage. The same toolkit now fits any organization sitting on years of accumulated data with a small team.
The ground rules never changed. Originals are never modified. AI suggests, scripts execute, a person approves. Client material stays on hardware I control. Every step leaves a record. And the structure underneath is older than the tools: I drew its outline in 2005.
Go deeper: start where you are
This page is the overview. Behind it is a set of pages with the numbers, the method and the mistakes, organized by what you want to know.
If you have 60 seconds
- Executive summary: who I am, what I bring, what one project proved
- Work with me: pick the path that sounds like you
If you want to know what I can do
- Technical skills: storage, networking, data, video, AI operations
- Types of projects: what I would do if I walked into your environment
- Specific projects: from family video to a 126 TiB archive
- Six AI skill areas: guardrails, agents, LLMOps, evals, retrieval, fine-tuning, with honest status
If you want to see the work
- The journey: thirteen phases in three stages
- Hardware limits: eight machines and what each one limited
- The copy map: what 14.7 million fingerprints revealed
- Two nights: truncated copies and flat 360 videos
If you want to know how I run it
- Guardrails: the risk ladder and the rules that made walking away safe
- Field lessons: twelve things that changed the next job
- Client playbook: the six-step engagement and weekly scorecard
- Results ledger: every number with its source
How I run it: an AI team, with a human in charge
The work runs as a small team. A manager agent plans and checks the work, two worker agents do the heavy lifting in containers, and local GPU models handle the bulk jobs, so nothing leaves my hardware and nothing costs a model call. I am the owner and the approver. The agents plan, execute and verify, and anything destructive waits for my sign-off.
The rules are mine, and they are simple. Every approval is logged word for word. Evidence, or it did not happen. Stop when something is ambiguous. Test on a copy before production. One morning showed what that buys: about 30 minutes of my decisions, and the team emptied and verified a failing 1.5 TB drive, close to 119,000 files, with every one of them checked.
Video projects
Local AI agents inventory, sort and tag large collections of footage so a search for “aerial, church, autumn” returns clips instead of folders. Footage is fingerprinted and verified before anything is touched. Read how the golden record works.
Video creation
AI-assisted shortlisting takes a mountain of drone and camera clips and surfaces the strongest candidates for a human editor, who makes the creative calls. Faster rough cuts, more time on story. See how I approach drone storytelling.
Massive file store management
A TrueNAS-based archive acts as the master library, with verified copies, checksums and a plan for safe compression that reclaims working space while preserving the raw sources for future editing and production. The archival compression approach.
Hardware and network improvements
Measure first, then upgrade the part that is actually holding you back. Segmented networks, scripted checks and safe, reversible change. Getting the most from the hardware you have and deploying network change faster and safer.
SCAN AI
ScanAI is a live capture and transcription system I built for local public-safety radio traffic. It records each transmission, transcribes it, scores clip quality, tracks activity by town and by hour, and monitors itself with a watchdog so it keeps running unattended. It currently holds more than 15,000 clips from six local towns.
It is a private tool, and access is by request only. Read how ScanAI works, then contact me if you would like a look.
What it means for you
If you are an executive or HR leader weighing AI development and solutions, this is what I can show you: real systems, running every day, built with sensible limits. Messy data made searchable. Slow manual processes made repeatable. And a person still accountable for the outcome.