AI Lab
kenneth is a portfolio with a small, bounded AI assistant. These pages explain how it is built, how it is tested, and where it falls short. Numbers are labeled as measured, scripted, or target.
Where things stand
- Scripted evaluation40 of 40report-set cases pass
Fixture evaluation, 2026-10-22, with a scripted model. This tests the server code around the model.
Live-model evaluation: not run.
How well the live model behaves is not yet measured.
- Production reliability: no production data exists.
Inside the assistant
4 articlesReviewed
Architecture
How the kenneth assistant is built: the request path, grounding in public content, and where each trust boundary sits.
Read articleReviewed
Threat model
The prompt-injection and privacy threats the assistant is designed against, the control for each, and what is still open.
Read articleReviewed
Evaluation
How the assistant is tested, what the numbers mean, what has and has not been measured, and the failures found along the way.
Read articleReviewed
Cost, latency, and tradeoffs
What the assistant costs to run, how fast it was observed to be, and the choices made to keep both small.
Read article
How to read the labels
- Measured
- A number from a saved evaluation run, with its denominator.
- Scripted
- A number from a run where a script stood in for the model.
- Target
- A threshold set in advance. It is not a result.
- Not run
- Nothing has been measured yet.