Work

Recent work.

Two recent engagements in different industries, with one approach: map how the business works, then build on top of it without breaking it.

CONSTRUCTION · JUNE TO OCTOBER 2026

A construction company

Quotes, estimates, schedules and crew sheets lived across shared files. The goal was an AI layer on top of them without changing how the team works.

430quotes queryable

4,700crew-day assignments

13ChatGPT tools

0writes to source files

01A read-only sync copies the company's files into a database every two hours. The originals stay the system of record and are never written to.
02A hub app shows job status, the schedule, the crew week and money in. Questions the system can't answer on its own go to the team, and their answers become rules.
03A ChatGPT connector answers where a job stands, who's on site, similar past jobs and pricing, and drafts estimates and quotes on the company's own template.
04Nightly checks reconcile notes against the schedules and raise flags. Sensitive data like HR and payroll is excluded.

MEDIA · EARLY TO OCTOBER 2026

Dexerto

A global gaming and entertainment publisher. Sales and campaign teams were quoting conflicting numbers from scattered sources. The goal was answers anyone could trust.

0.94answer accuracy, up from 0.68

112kDrive files catalogued

~20active users on the assistant

0invented numbers

01Started with Dexpert, a Slack assistant answering sales, account and strategy questions from company data, tied into HubSpot, Asana and Drive.
02Built the audience and research tools around it, including one agreed audience size so every pitch quotes the same numbers.
03Rebuilt the data layer into one database with curated views and context docs, so any colleague can ask Claude and get sourced answers.
04An evaluation system tests answers against the decks the team actually shipped, every day.

Not sure where to start? Most clients aren't.

Book a free call