Notes on legal AI, written by people who have to make it work
Why we reject findings that fail the exact character offset test
Most legal AI products work on summaries of chunks. Here is what that gets wrong in an indemnities review, and why we anchored CitationGate to verbatim spans instead.
Building a house playbook when your partners disagree on everything
A method for turning a shelf of closed matters into one tiered position document a partnership can actually live with, and what to do about the clauses nobody agrees on.
The Nigeria Data Protection Act (NDPA) 2023, the NDPR and cross-border legal documents: what the regulation actually requires
The parts of Nigerian data-protection law that matter to legal ops teams: which instrument is the primary statute, what counts as a cross-border transfer, and where an AI review fits.
The Lawyer Analysis Metric: documenting human supervision in-tenant
Why the audit trail needs to record who approved a finding, who overruled it, and under which supervision policy, and what that record is for.
Writing native Word revisions from a web task pane without corrupting styles
Why a redline has to arrive as real tracked changes in a real .docx, and the document-model problems that stand between a web app and that file.
Per-seat pricing is broken for legal AI, and here is the math
Why charging per lawyer makes firms hoard licences and kills adoption, and what a credit model has to get right to be better rather than merely different.
The sovereignty tier: running local inference for banking and government briefs
Hardware, latency and the tradeoffs of running the whole stack inside your own network, including the one requirement people try to skip.