Portfolio
Data Analyst & AI Specialist
I turn messy, real-world data into tools people actually use. First Class Honours in Mathematics and Computer Science from Lancaster University, plus two years building data products and AI-driven automation for schools, charities and local authorities, from dashboards that opened new business to processes that cut hours of manual work down to minutes.
Projects from my time as a data analyst and freelance AI specialist. Client names are kept general, ask me for more detail.
A three-stage AI workflow that turns raw data into a structured source of truth, writes commentary in the client's own voice, then formats it into their branded Word template. Generic AI output wasn't good enough, it had to read like the client wrote it themselves. A process that used to take several hours now takes four to five minutes, ready for a light review rather than a rewrite.
Built a tool that generates a full set of tailored surveys straight from a client's own governance framework, carrying forward anything left incomplete from the previous term, then feeds the responses into a live dashboard tracking status by group. Led to two further commissions from the same client.
Designed a composite "Inclusion Index" from public school-level data on attendance, exclusions, demographics, attainment and SEND provision, aggregated to Trust level and built into a public, interactive Tableau dashboard with anonymous peer benchmarking. Owned it end to end, from metric design through to data sourcing, build, QA and deployment.
Redesigned a client's data collection form, reprocessed years of messy survey data into the new structure, and built a guided Excel workbook that walks a semi-technical user through cleaning and exporting it, with VBA macros auto-filling gaps via the Companies House and Postcodes.io APIs. Also rebuilt the Tableau dashboard it feeds, from scratch. What used to be a full-time job became a 20-minute task, every two weeks.
A research report on SEND provision gaps across London led to one of its own recommendations being commissioned as a live tool. Sourced and joined public datasets, packaged them as a documented Excel tool for the client's team to maintain themselves, and built an interactive Power BI map with filtering and tooltips on top. Now live on the client's website.
Things I've built outside of client work to understand how they work under the hood. Newest first, and the full running list lives on the blog.
A from-first-principles look at the Perlin noise algorithm behind games like Minecraft, covering octaves and Fractal Brownian Motion, with a working 2D and 3D terrain implementation. Read the write-up →
Implemented K-Means clustering from scratch in NumPy for image segmentation and colour reduction. Read the write-up →
Built a genetic algorithm from scratch using selection, crossover and mutation, beating a greedy-algorithm baseline by 43%. Read the write-up →
Built roughly 60% of the codebase for a full graphics editor, covering the brush engine, layers, transforms, undo/redo, a UML-designed architecture, and real usability testing with live participants.
An A-Level project. It features a 3D cube you can manipulate, an optimal solver, and a hand-built tutorial teaching the beginner's method step by step, rendered in OpenGL with a Tkinter interface.
First Class Honours
Strongest modules included Artificial Intelligence, Real Analysis, Databases, Graph Theory, Security & Risk and Computational Mathematics, all graded first class. No dissertation. Coursework included a group project (Paint++, above) instead.
Computer Science (A), Mathematics (A), Further Mathematics (A*), Chemistry (A)
7 grade 9s, including English, Mathematics and Science, plus Further Mathematics (A*)
jdhwilkins.com is a project I've kept building since August 2024, evolving from programming and data-analysis write-ups into experiments and lessons from using AI day to day. I designed and built the site myself on Ghost.
I've written close to 50 pieces, published here and on Medium, building a following of over 5,400 readers and around 4,100 email subscribers on Medium. My highest-performing piece, a guide to prompt engineering, has been viewed almost 300,000 times on Medium alone.
If any of this is relevant to something you're working on, or you'd just like to talk data, AI or writing, get in touch.
