AI Transformation / Legal Operations
A decade in practice, from litigation to international tax, then a move into the engine room: AI drafting workflows, search across confidential document sets, automated review pipelines, and the training that decides whether any of it actually gets used.
I look for the seam where legal judgment meets repetitive work, remove the repetition, and measure what changed.
Putting an LLM in front of lawyers
Several months building an AI drafting assistant inside a law firm, then rolling it out across the firm. Keeping client names away from the model entirely, teaching it to stop and ask instead of guessing, and the week it spent printing its own instructions at the top of every letter.
A working reference implementation of the confidentiality design from the essay: access control enforced in the data layer rather than the application, so no mistake in the application can widen who sees what. Infrastructure, search configuration, answer generation with citations, and a harness for measuring retrieval quality. Built on Azure, parameterized rather than tied to any tenant.