Work
Systems we put into production
Written up the way we would want to read them: what was broken, what we did, what we would do differently, and what it changed. These are illustrative engagements — our first public client references are still in build.
Rebuilding a payments ledger without stopping payments
A single-table balance model was drifting under partial refunds. We moved it to double-entry behind a shadow-write, then cut over with zero downtime.
- 0Minutes of downtime
- 11Silent drift bugs found
- 3.2MEntries backfilled
A voice agent that survived a 400-room property
Front desk was losing a third of inbound calls at peak. We put a voice agent in front of the queue that books, confirms, and escalates — with every turn transcribed.
- 31%Calls previously abandoned
- <800msMedian turn latency
- 68%Handled without a human
Turning audit prep from six weeks into a report
Evidence lived in email, shared drives, and three people's heads. We built an append-only evidence store mapped to their control framework.
- 6 wks → 2 daysAudit preparation time
- 100%Controls with linked evidence
- 0Mutable records
Next step
Have something similar?
Most of these started as a conversation about a number that would not reconcile.
Replies within one working day