Skip to content
Former lawyer · AI for regulated firms

AI for law firms and deal teams, private by design.

Adoption, deal sourcing, contract drafting and secure workflows. Every system checked before it is trusted.

System architecture Live model
01Matter input
05 linked stages02 review gates01 governed system

Regulated work can't run on borrowed tools.

Client data stays where the firm can see it, every output points to its source, and a person signs off. Why that comes first.

Working model

A matter is a system. Make every decision visible.

Active stage

Diagnose

Map the actual matter workflow before choosing technology.

System signalInput → decision → output
  • Task boundary
  • Human sign-off
  • Cost of error
The difference

Accuracy is an evidence question, not a vendor claim.

Every system published here shows its architecture, the evidence available, the evidence still missing and the failure modes that matter.

Worked example

How a threshold gets set.

Performance is scored against verified answers before it enters a claim. Hover a point to read the pair it came from.

Name-match threshold, measured
Hover a point to read the pair
Genuine matches scored 76–100, genuine mismatches 34–38. The threshold sits at 60, in the empty gap, set low on purpose: a false alarm on a legitimately shortened caption is how a tool gets switched off.

Bring the workflow that needs to change.

Start a conversation