Personal AI projects & experiments
A personal space for exploring what modern AI can actually do in the real world.
Not a commercial service. Just a genuine, ongoing curiosity—followed far enough to become something you can see.
Begin with the workSelected explorations
The examples carry the detail. Each one starts with a practical question, follows the decisions behind it, and shows its boundaries as clearly as its result.
Existing public mock-up
The idea began with a familiar place and a broad question. Instead of answering with a feature list, the project turned that question into a suite of tangible concepts—from marine-life identification to dive planning and underwater photography.
The useful part Making an unfamiliar technology understandable through situations people already recognise.
Read the Dive Gurus storyClean-room recreation
A fulfilment journey becomes trustworthy only when success, refusal and unchanged state are equally clear. This recreation follows four fictional requests through those moments.
The useful part Turning hidden process rules into feedback a person can act on.
Read the fulfilment storyClean-room recreation
Support teams often know what they need to find but not how a system expects them to ask. This example translates an ordinary question into a readable sample filter.
The useful part Keeping technical logic visible enough to check and simple enough to reuse.
Read the query storyClean-room recreation
A useful assistant should recognise uncertainty rather than disguise it. Five fictional media items move through inspection, identification and planning before a human decides what happens next.
The useful part Treating ambiguity, approval and audit history as features—not friction.
Read the Median storyClean-room recreation
A browser-only simulation makes the invisible safeguards visible: a bounded request, a staged preview, human approval and a way back.
The useful part Designing confidence and recovery into the path, not around it.
Read the publishing storyClean-room recreation
This fictional journal story follows an idea from conversation through structure, consistency checks and a safe reject-and-restore moment.
The useful part Preserving a human voice while making each hand-off easier to understand.
Read the journal storyClean-room recreation
A fictional journey record supports an itinerary, route, status table and poster while keeping offline limits and missing information visible.
The useful part Letting presentation change without allowing the underlying story to drift.
Read the journey storyThe thread
A quiet boundary
Local experiments are kept separate from corporate data. Public demonstrations use fictional or public material, remain honest about what is simulated, and include no employer, client, family or friend information.
At the back of the book
David Halliday Switzerland
Hi, I’m David. My background is in IT support and systems analysis; Divedog is where I follow practical questions into working experiments with automation and AI.
Some begin with service friction, others with publishing, media, travel or diving. The common thread is simple: understand the problem, make the moving parts clearer, and leave room for human judgment.
Say hello ai.divedog@gmail.com →