I'm a California litigation defense attorney who builds the tools instead of waiting for them. My practice is wage-and-hour, PAGA, FEHA, and consumer warranty work, which is high-volume and pattern-heavy, exactly where good engineering pays off. I started building to solve my own problems. The first real system was a discovery-response drafter built on a selection-not-generation design: an attorney-curated objection library, deterministic rendering, and a verification pass, so the model selects from vetted firm language instead of inventing it. It's production-grade and heavily tested, built clean on synthetic and de-identified data. From there I've been working through the rest of the litigation lifecycle, treating each repetitive task as a candidate for the same propose, review, commit pattern. What I care about most is the governance layer: privilege handling, conflicts checks, tamper-evident audit trails, prompt-injection defense, and a fail-closed de-identification pipeline. The hard part of legal AI isn't the drafting. It's making the output trustworthy enough to put in front of a partner and defensible enough to stand behind. I'm an IAPP AIGP and a LegalQuants resident, and I'm most interested in talking with other people shipping real litigation tooling.