The first six are work I have delivered, described without client names. The last three are example scenarios: they show how I would approach a typical problem and are not client projects.

Filter by sector
Delivered work

Turning supplier emails into structured quotes

  • Aviation
  • Documents
  • LLMs

An aviation spare-parts company received supplier quotes as free-form emails. I built a system that reads their mailboxes and turns each real quote into a structured record. I built and integrated it alone.

Problem

Prices and part details were buried in PDF, Word, Excel or ZIP attachments, so buyers had to open, read and re-type every quote.

Approach

The system reads the mailboxes, recognises real quotes and extracts part numbers, condition codes, prices and certificates into the procurement platform already in use. Follow-up replies update the same record, failures are retried, and buyers can review and correct every result.

Result

  • In daily use
  • Much improved how the team processes and searches supplier quotes

Relevant if your team re-types quotes, orders, invoices or shipping documents.

Delivered work

Deciding on board which satellite images are worth sending

  • Space
  • Vision
  • Edge devices

For a company building embedded space systems, I designed a filtering system that decides which images to keep and send down.

Problem

A satellite has limited compute and bandwidth, so it cannot send every image. The choice has to be made on board, and accurately enough to be worth the energy.

Approach

The system balances accuracy against the compute and bandwidth available, and can learn new targets from a few examples. The same models run on servers or on embedded hardware.

Result

  • A working version is complete and is being improved
  • A much better balance between accuracy and the data and energy budget

Relevant if AI has to make decisions on limited hardware, such as cameras on a production line.

Delivered work

RetroBench: a test AI vendors can't game

  • Technology
  • Evaluation
  • LLMs

RetroBench tests whether AI coding agents can rebuild source code that compiles byte for byte to original 1990s game code.

Problem

Public benchmark tasks end up in training data, so a high score can mean the model remembers the answer and not that it can do the work.

Approach

The tasks ask for source code that compiles to exactly the original bytes. No model can have memorised them, and the result is checked by the compiler, not by opinion.

Result

  • 15 models from the major labs tested
  • Scores range from about 27% to 98%

Relevant if you are choosing between AI models or vendors: I build the same kind of test from your own cases.

Delivered work

Demand and supply modelling across sports for a national programme

  • Sport
  • Forecasting
  • Data

For a national programme, delivered with a UK partner, I built models that answer a planning question from the data that could actually be found.

Problem

The programme needed estimates of supply for different sports, both overall and by region. The inputs were demographic and statistical data, survey data and research findings, and none of them covered the question completely.

Approach

I built the models to fit the data that was available, with assumptions that can be explained in plain words and changed easily.

Result

  • Working models delivered
  • Assumptions are explicit and easy to change

Relevant if you have a planning question and partial data, but no one to model it.

Delivered work

A multi-agent research platform

  • Technology
  • Agents
  • LLMs

A research platform with a critic agent, a full audit log and about 3,300 automated tests.

Problem

A language model with a search tool usually stays within the keywords it chose first and the first perspective it took, so its research is narrow.

Approach

An orchestrator agent splits the work into focused tasks for several agents, which search iteratively. The agents share what they find, so one agent's results change where the others look and which questions they answer. The research changes direction as knowledge builds up, as a real research process does, and ends in a report.

Result

  • A critic agent reviews the work
  • Every step is recorded in a full audit log
  • About 3,300 automated tests

Relevant if you need AI agents whose work can be checked afterwards.

Illustrative scenario First test: 1–3 weeks

Fraud detection for a financial institution

  • Finance
  • Fraud detection
  • Data protection

An example of a first project for a bank or payment company that wants to improve its fraud detection and has to keep personal data protected.

Problem

Fraud patterns change faster than hand-written rules, and every false alarm costs an analyst time and annoys a real customer. Transaction data is personal data, so it has to stay inside the institution.

Approach

I would start from your own labelled history, inside your environment. The first step is a test set and a baseline for the current rules; then anomaly and behaviour models are compared with that baseline on the same cases.

What a first test would measure

  • How much fraud the current system catches and misses
  • How many alerts are false
  • Whether a model beats that baseline on your own data
Illustrative scenario First test: 1–3 weeks

Predictive maintenance in manufacturing

  • Manufacturing
  • Predictive maintenance
  • Sensor data

An example of a first project for a manufacturer that records machine data and still learns about failures when the line stops.

Problem

Unplanned stops are expensive, but so is replacing parts too early. Whether the sensor data contains a usable early signal is not known until it is tested on past failures.

Approach

I would take the failure and maintenance records you already have, build a test set from past stops, and measure how early and how reliably a model would have warned.

What a first test would measure

  • How many past failures would have been flagged in advance
  • How much warning time there would have been
  • How many warnings would have been false
Illustrative scenario First test: 1–3 weeks

Customer churn for a telecom operator

  • Telecommunications
  • Churn analysis
  • Forecasting

An example of a first project for an operator that sees customers leave and cannot tell in advance which ones.

Problem

Retention offers cost money, so they should go to customers who would otherwise leave. Without a tested model, offers go to too many people or to the wrong ones.

Approach

I would build a test set from past months of usage, billing and support records, and measure how well a model ranks customers by their risk of leaving, compared with the rule in use today.

What a first test would measure

  • How many of the customers who left would have been in the top risk group
  • How that compares with the current targeting
  • Which signals carry the prediction

Ready to Transform Your Business with AI?

Book a free intro call with me to discuss your business challenges and opportunities.

Book a Free Intro Call