AI Projects › Executive summary
If you are sitting on years of accumulated data, hardware and half-finished systems, I can walk into that environment, work out what you actually have, and make it safe, searchable and worth more.
I have spent my career on the fence between sales engineering and marketing: the person who understands the machine and also knows what the customer needs to hear. Decades of marketing and sales leadership taught me which information is valuable enough to mine. Ten years inside warehouses and equipment environments taught me how things really break. I am still willing to do the hand work: carry the drive, trace the cable, read the log.
What I bring
| Strength | What it looks like in practice |
|---|---|
| Marketing and sales leadership | I know which data tells a story a buyer will act on, and which is noise. |
| Mechanical and hardware fluency | Warehouse and equipment environments, plus a home data center I built and run myself. |
| Data hardware, software and databases | Storage servers, networks, Postgres, GPU video processing, and the scripts that tie them together. |
| AI used as a tool, human in charge | A small team of AI agents does the tedious work. Nothing destructive happens without a written approval. |
| Video and drone | A decade or more of production, so I understand the assets, not just the storage. |
What a recent project proved
Over roughly 45 days, working about 12 hours a day, I put back together a video library that had been copied forward across about twenty generations of drives. Every number below comes from the project records and was current on November 10, 2026.
| Result | Number | Why it matters |
|---|---|---|
| Fleet fingerprinted | 126 TiB, 14.7 million files | Every later decision is provable |
| Single-copy exposure ranked | 12.9 TiB | One disk failure from loss, so protected first |
| Redundancy quantified | 82.6 TiB theoretical | Reclaimed only where two other proven copies exist |
| Compression on passing files | 5.69 TB to 0.83 TB | About 85% smaller, each file linked to its original |
| Silent corruption caught | 4 truncated camera originals | A name-and-size check would have passed them |
| Failing drive emptied unattended | 118,599 files, 1.52 TB | Counts, bytes and dates all matched |
| Unapproved destructive actions | 0 | Nothing deleted without a proven copy and written approval |
Symptoms lie
I do not manipulate data. I look at the problem for what it is, recognize the symptoms, and treat the disease. Four times in one project, the first diagnosis was wrong:
- “A dying server.” Three hard resets pointed at failing hardware. Console photos showed a corrupt pool panicking at boot. The reformat was cancelled and the data was salvaged.
- “A slow server.” The 10 gigabit machine was fine. A 1 gigabit mini-PC set the speed of everything behind it.
- “A good video.” It scored 96 on quality and still played flat, because a metadata tag was gone.
- “Done” migrations. The labels said complete. The data said 6 to 24 percent.
Each one was found by checking the result against the source, not the tool’s own report. Read the two nights and the field lessons for the detail.
Why it matters to you
The same approach works on a marketing department’s media library, an agency’s old drives, a manufacturer’s shared folders, or a small organization whose data simply grew out of control. The goal is always the same: know what you have, protect what matters, reclaim what is wasted, and make the rest findable.
Where to go next
- Types of projects: what I would do if I walked into your environment.
- Specific projects: the real work, from personal to professional.
- The client playbook: the six-step engagement.
- The results ledger: every number with its source.
- Technical skills: the hands-on list.
Figures are from the project fingerprint report (Nov 7), lineage and encode reports (Nov 8 to 9) and the decisions log. Still-moving items are marked on the results ledger.