Work and prototypes

Things I have built, and what they show.

My own builds, run every day, and work from my time at Yodel. Each started with a real problem. AI was the toolset, not the point.

A cobalt ribbon tied in a loose knot, flowing into a smooth wave.

Running every day

My own AI workspace

The set-up I run my own work on, and the example behind every AI workspace I build.

My plans, projects and processes live in one notes vault, with one page that holds my current plan and priorities.

The vault is the second brain. Claude works from it when I build, write and think, and every assistant shares the same skills, tool keys and rules. Background assistants run job and contract searches, site checks and a daily data log on a schedule, and email me a summary. Every scheduled run sends a message, even when nothing is new, so silence means something is broken.

Two rules keep it safe: assistants ask me before changing the plan, and nothing goes to other people without my OK.

Built with: Obsidian, Claude, scheduled AI assistants, GitHub and Vercel
Shows: One source of truth that several AI tools work from, with human approval built in

Live

Design Your Prompt

An online tool for building multi-step AI prompt workflows, testing them with a team and turning the proven ones into Claude Skills.

I first built it so my son could run a chain of prompts to research, design, build and market a website, with AI at every step. It grew into a full prompt library.

Workflows run one stage at a time. Variables fill in the details for each client or project. Sensitive-data rules anonymise information before it goes to the AI and put it back in the response. Teams share workflows with folders and tags, and a tested workflow exports as a Claude Skill in one step.

Built with: Claude Code, using an agent team with product manager, sprint, build, test and release roles
Shows: A real product built and shipped with AI, for the problem of prompts kept in Word files

Visit Design Your Prompt

Paper testing

ValueBetEdge

My racing data and forecasting project, and the place I practise modelling.

It has my own database of historic and upcoming races, a daily automated data download, and models tested against Betfair Starting Price, the market’s closing price.

Straight betting strategies have not beaten Starting Price so far, so they are not used with real money. That is the point of testing properly before you trust a model.

Now a paper-only test of a new selection idea runs every day. A background assistant logs each selection and settles it at Starting Price, so the result builds up without any risk.

Built with: My own racing database, an automated daily download, Claude for the modelling and a scheduled assistant
Shows: Forecasting discipline: a clear pass mark, an honest test and no shortcuts

Yodel

Tableau to Power BI, with AI

AI-assisted workflows that sped up a dashboard migration inside a large company.

At Yodel I designed Claude-assisted workflows connected to the Teradata data warehouse, to speed up moving our Tableau dashboards to Power BI.

The reporting standards, the link back to the source data and human review all stayed in place.

Built with: Claude, Teradata, Tableau and Power BI
Shows: AI used on real company data, inside the company’s rules

Analytics at Yodel

Three problems. Three working answers.

All three at Yodel, a UK parcel carrier. Each went past the dashboard to something the business could act on.

Fraud detection

Stopping a fraud, not just reporting it

The head of security found a fraud and asked for a dashboard. I sized the problem, worked out how to spot it early and built alerts that tracked how it changed.

Security officers used the alerts to intercept parcel collections. The intelligence supported a criminal case and was shared with other UK carriers.

Location planning

Where to put the next collection point

Each round of candidate stores took the old process two weeks, and its software had broken. I rebuilt it with the director of stores in Python, with interactive mapping.

The new process weighed population, nearby stores, delivery volumes and store feedback, then assessed and recommended sites instead of showing a map of dots. The same approach helped place out-of-home advertising locations.

Volume forecasting

More detailed, more accurate parcel forecasts

I built forecasting models in Azure Machine Learning that added detail and accuracy to the group’s parcel volume forecasts.

I also developed forecasting models with InPost’s Data Science and AI team after InPost bought Yodel.

Recommendation

In my former line manager’s own words.

Former line manager, Yodel. LinkedIn recommendation, October 2026. See it on LinkedIn.

I had the pleasure of working with James from 2015, when he joined the organisation to lead the UK Data Team. It is rare to find someone who combines such strong leadership and organisational skills with a genuine passion for learning, embracing new technologies and putting them into practice. James brings all of these qualities together exceptionally well.

He was a fantastic coach to his team and a highly respected project leader, earning the trust of colleagues through his ability to bring people together, develop capability and turn ideas into meaningful outcomes. His well-deserved promotion to lead a new Analytics and AI function was a natural progression.

In this role, James led the design and delivery of our new AI platform, delivering significant commercial benefits across the business. These included preventing fraud, optimising the placement of Out of Home advertising locations, and enhancing the granularity and accuracy of the group’s volume forecasting models.

James was highly regarded well beyond his own team and recognised as a leader in analytics and AI. What sets him apart is his ability to combine technical knowledge with an understanding of the wider business, alongside a collaborative approach that makes him an excellent sounding board for ideas well beyond his own area of expertise.

James was a pleasure to work with, and I personally miss his insight, challenge and contribution. I would wholeheartedly recommend him to any organisation looking for an accomplished analytics and AI leader who can combine strategic thinking, technical innovation and outstanding people leadership to deliver real business value.

Get in touch

Want something like this for your work?

Start with a One-Hour AI Build on one task of your own, or tell me about a bigger problem.

Get in touch