Work that shipped.
One paragraph each.

Dallas's original small-batch roaster was running wholesale orders, roast schedules, and green inventory across three spreadsheets and a shared inbox. We put all of it in one place, generated the roast plan from real orders, and printed roast-date labels straight from the batch log. The owners stopped reconciling and went back to roasting.

A seventy-year-old venue reopened with a full concert calendar and five systems that had to agree about every show. Now each show is entered once and flows everywhere, ticket buyers and email subscribers live in one audience, and settlement runs the morning after instead of the week after. The team spends its time on shows.

A 24-story Turtle Creek condominium with a round-the-clock front desk and a lot of binders. We moved packages, visitors, vendors, and maintenance requests onto a tablet and a resident portal, consolidated the governing documents, and put an assistant on top that answers "can I put a grill on my balcony" the same way no matter who is on shift.

An independent operator growing by acquisition, which means every deal arrives with its own well files, leases, and accounting quirks. We joined production, land, and financials at the well level, made management reporting refresh daily, and indexed the lease agreements so the land team gets a first answer in minutes rather than an afternoon.

Trauma and emergency-response instructors who were spending their weeks on scheduling, rosters, and certification reminders instead of teaching. We built the course operations they needed, automated the renewals, and added a conversational scenario trainer so students can rehearse decisions between live sessions. More teaching, less admin.

TheRecordXchange produces the official court record, so "the AI transcription is pretty good" is not an acceptable answer. We built an evaluation harness on real courtroom audio, reviewed the record and delivery pipeline end to end, and turned model and vendor choices into measured decisions with numbers attached.

Millions of streams analyzed daily on oversized servers, deployed by hand, with an AWS bill nobody could explain. We moved compute to autoscaling containers, split analytical reads off the transactional database, replaced shell scripts with real pipelines, and tagged every dollar. Dashboard latency dropped 60%, the bill dropped 38%, and deploys went from weekly to daily.