Chief Technology Officer

An early-stage team without a real process: no shared cadence, unclear priorities, and knowledge locked in individual heads. The rest of the company could not see what engineering was doing.

Give the team a rhythm that produces predictable delivery and visible progress.

Set up two-week sprints with grooming, solutioning, and retro meetings, and instituted regular knowledge transfer sessions within the team.

Team happiness went up 38%, and for the first time, the company actually knew what the engineers were working on.

The team had started using AI for coding, but loosely: everyone improvising, no shared discipline, inconsistent results.

Turn AI-assisted development from a personal habit into an engineering capability

Led the shift to agentic programming and built a strong shared harness that every developer uses, running on Claude Code and fully integrated with Jira, Datadog, and our other systems.

Development speeds up 4x with 28% fewer bugs. Faster and better, not faster and sloppier.

Workflows across the organization were manual, and every person who tried AI used it their own way.

Move the whole organization from manual to AI-informed, and where it proves itself, AI-actioned.

Integrated Claude deeply into the organization, built a library of custom skills managed in a shared repo so everyone works with AI the same way, and trained people on what exists and what it does for their daily work.

One consistent AI way of working across the company instead of scattered experiments.

In automotive parts, finding the right part is a solved problem with external data providers. In off-highway parts, those providers do not exist, and the data is not available anywhere.

Make parts findable in a market where the data to do so had to be created first.

Built our own partfinder dataset using scraping and machine learning models, then built the partfinder on top of it.

A proprietary database covering over 60,000 vehicle types, in a market where external providers offer nothing. Recently released; conversion numbers to follow.

Our way of working is too specific for off-the-shelf order management, but a custom OMS is normally an 8 to 12 month project.

Build an OMS designed for exactly how this business operates, fast.

Built it fully in-house with two full-time developers, using our agentic engineering setup.

Delivered in two months. Customer reviews went from 3.6 to 4.2 stars in the two months after launch.

Category and product pages took 2.45 seconds to respond in Magento, and relevant products were drowned out in the results. The usual fix is buying Algolia, Fredhopper, or Rich Relevance.

Make search fast and results relevant, without the external bill.

Built custom middleware connecting product indexing and API calls directly to OpenSearch, plus our own sorting logic based on sales, data completeness, and other signals. Agentic coding made building this in-house fast.

Response times under 300 ms, more relevant products on top, and roughly $200k a year saved on external search tooling.

Decisions ran on guessing: data was scattered across systems, and the people who could read it were not the people who needed it.

Put the data in one place and make it answerable by anyone.

Built a data lake with bronze, silver, and gold layers, then connected a Slack agent to Claude Code CLI so it queries the database directly.

Anyone can ask any data question in Slack at any moment and get an answer. Decisions moved from guessing to data.