Harbor Ledger
A private model trained on ten years of internal memos and term sheets. Used for first-pass review, never for the signature.
AI studio · Toronto
An AI studio for operators who cannot afford drift. Strategy first, then custom models and agents that stay on course.
01 — Practice
Three disciplines, one constraint: no system ships without a named owner and a way back to the brief.
The brief before anyone trains a thing. Scope, risk, and the decision the model is actually there to support.
Custom weights on your own corpus. Fine-tunes and retrieval that stay inside the walls you already trust.
Live loops with a person still in them. Tools, hand-offs, and a kill switch that is not decorative.
02 — Selected work
A short list. Names are changed; the constraint is not.
A private model trained on ten years of internal memos and term sheets. Used for first-pass review, never for the signature.
An agent loop that drafts dispatch and flags exceptions. A supervisor still closes every ticket that leaves the yard.
Six weeks on where not to apply a model. The outcome was a smaller build and a clearer owner on the floor.
03 — Method
Map the constraint, the owner, and the decision that must not drift.
Gather the documents, tools, and exceptions the system has to know by name.
Train or compose on that material. Measure against the brief, not a public leaderboard.
Ship with a person in the loop and a review cadence that survives the first quiet month.
“They would not let us skip the brief. That is why the model still matches the work six months later.”
04 — Write
Tell us the decision that cannot drift. We read every note.