AI Projects › Types of projects
If I walked into your environment, I would not start by buying anything. I would work out what you already have, what is at risk, and where a little money moves the needle most.
Network topology and security
Where does traffic really flow, what is exposed, and what is the slowest link? In my own work the obvious suspect was the 10 gigabit server. The real bottleneck was a 1 gigabit mini-PC. Measuring first kept the fix cheap and aimed at the right machine. Public access was designed as one front door, not a pile of forwards, after logged intrusion attempts on an old one.
What you get: a map of the network as it really is, a list of exposures, and a ranked fix list.
Hardware optimization
Old equipment usually has more life in it than people think, and a specific limit that shapes the plan. I map each machine, name what it limits, and design around it: the right job on the right box at the right hour. See the eight machines and their limits.
What you get: a schedule set by your slowest device and a short list of upgrades that actually move the finish line.
Investing in the right places
Every dollar of upgrade should be tied to a measured constraint. A ranked list of bottlenecks and single points of failure turns “my storage is a mess” into a work order with costs attached. See the copy map for how one chart did that.
What you get: spending aimed at the constraint, with before-and-after numbers.
Data inventory, protection and cleanup
Fingerprint everything, find what exists only once, protect it, then reclaim only what is proven redundant. Nothing is deleted on a guess.
What you get: an inventory you can trust, a protect-first list, and a ledger of every removal.
Media and marketing assets
Footage, photos and creative files are some of the most valuable and most neglected data a company owns. I know how to find the pieces that matter to marketing and sales, and how to keep their metadata intact.
What you get: a library a person can search, with originals untouched.
AI-assisted operations
A manager agent plans and verifies, worker agents execute, and a person approves anything that touches data. See the risk ladder.
What you get: work that continues while you are away, with a paper trail.
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.
The first engagement
I identify before I act: read-only access, a sized map of the estate in the first week, a schedule set by the slowest device, and written rules about who can approve what. Then we move. The full sequence is in the client playbook.