Demos / pk aidoo

Two products you can actually click into.

These are the real applications, deployed separately and seeded with invented data, so you can use the features instead of looking at screenshots of them. Nothing here is connected to a client, and nothing asks you to sign up.

Sandbox environments. Sample data, not connected to production.

Every company, person, and document you will see is fabricated for the demo, including the studios and the newsroom. Anything you type stays in the sandbox and is cleared periodically. No account, no email, no signup.

Enso

A multi-tenant AI drive: documents file themselves, and chat answers questions about them with citations.

Things worth trying

  • Open a document and look at the folder it sits in. The drive worked that grouping out on its own, rather than someone filing it there.
  • Ask the chat something that spans two or three documents, then click a citation and check it points at the passage the answer came from.
  • Vote a bad answer down. Retrieval feedback is a live signal in the real product, not decoration.

React, Vercel functions, Supabase with row-level security, Postgres full-text retrieval, Claude for classification and grounded chat.

Open the Enso demo

Teshie Journal

An editorial publishing pipeline: draft, in review, returned, approved, published, with the review tools that go with it.

Things worth trying

  • Move a seeded draft through the pipeline, including sending one back to its writer with a reason they will actually read.
  • Compare two versions of a piece and read the diff, then leave an inline comment or a suggestion the writer can accept or reject.
  • Open the Filler desk, where drafted social posts are waiting on your review, each scored against the newsroom's own editorial rubric.

React, TipTap editing, Supabase, Claude for the content pipeline. Scheduling and email are stubbed in the sandbox.

Open the Journal demo

The AI features run for real. Classification, retrieval, and chat all call a live model, so answers are generated in the moment and will vary between visits rather than replaying a script. That also means they can be wrong, which is part of what there is to look at.

Because the calls are real, each sandbox has a daily cap. If the AI stops responding, the day's budget is spent, and it resets.