Get in touch

Applied AI Consultant · Solihull, UK

James Farrelly

I turn business problems into practical AI - scoped, shipped, and explained to the people who sign it off.

James Farrelly

Don’t just read a CV. Interrogate it.

Meet JamesGPT - trained on my real CV and answering as me. Ask it anything about my work; if it doesn’t know, it’ll point you to the best way to reach me. It’s the same kind of natural-language interface I build for clients.

JamesGPT · trained on my real CV, answering as me.

I gather the requirements, then build the thing myself - using AI to get there fast.

I work end to end: sit with the people who have the problem, pin down what they actually need, then go away and deliver it. No handoff, no long spec-to-developer relay - AI collapses that gap, so the time from “here’s the problem” to “here’s the working thing” is short.

My specialism is data engineering across Microsoft Fabric and the wider Azure stack - pipelines, models and reporting that hold up in production - with AI layered in where it genuinely earns its place.

Data privacy isn’t an afterthought either - I build with GDPR and data protection in mind from the start, not bolted on at the end.

Things I’ve built and shipped.

  1. 01

    Natural-language interface over Power BI - watch the demo on LinkedIn

    Plain-English chat over reporting data. Microsoft’s Power BI MCP server reads the semantic model, an Azure OpenAI model writes the DAX, and the answer comes back formatted - so non-technical people self-serve instead of queuing for an analyst. No Copilot licence, and it all stays inside your own Microsoft tenancy.

    • Azure OpenAI
    • Power BI MCP
    • DAX
    • React
    • Cosmos DB
    Watch the demo
    Watch the demo
  2. 02

    Information Management Toolkit

    A toolkit I built for the information management team at RLB Digital. Five tools in one app, each replacing a slow manual job - and each one built with AI in the loop rather than bolted on afterwards.

    • Python
    • FastAPI
    • React
    • IfcOpenShell
    • Azure
    See the tools

    COBieQC

    ~3 hrs a run

    Validates a contractor’s COBie asset spreadsheet against hundreds of rules for the right client template and RIBA stage, then hands back a marked-up workbook and a pass/fail report. It replaces a manual eight-step check that took about three hours.

    COBieQC: the run-validation screen, with template, project stage and a drop area for the workbook

    Verification Reports

    ~5 hrs a month

    Checks what a project has actually delivered and packages it into one Excel report, reading straight from a Morta hub or an Airtable base. It was a 30-minute job done at least ten times a month.

    Verification Reports: choosing a Morta hub or Airtable base to generate a report from

    IDS Check

    In progress

    The same idea as COBieQC but for the 3D model: check an IFC file against a buildingSMART IDS rulebook and get a pass/fail per rule. It also generates the rulebooks themselves. The prize here is replacing the very expensive third-party checkers the industry currently leans on.

    IDS Check: selecting a rulebook and dropping in an IFC model to validate

    Project Setup

    Fill the project details in once and it finds every placeholder across the project’s tables and documents and populates them, with a preview before anything is written and an undo afterwards.

    Project Setup: pasting a project's Morta hub URL to scan for placeholder values to fill in

    Contact Search

    Takes a project’s contact list, researches the companies behind it, and fills out the COBie Contact sheet properly instead of by hand.

    Contact Search: dropping in a COBie file or email list to fill in company, phone and address details
    5 tools, 1 app
  3. 03

    Unstructured spreadsheet → Power BI

    A messy client resource-and-cost workbook gets dropped in SharePoint; a flow hands it to a Python Azure Function that flattens the weekly grid into a clean, normalised table and lands it where Power BI reads it. Upload and walk away - no manual reformatting.

    • Azure Functions
    • Python
    • Power Automate
    • SharePoint
    • Power BI
    Upload → report
  4. 04

    Estate risk analytics for a major retailer

    Building-condition survey data enriched with free public environmental data - flood zones, subsidence, rainfall, coastal exposure - then scored so the highest-risk sites surface first. Built on Microsoft Fabric with Delta tables feeding an interactive dashboard. Ongoing.

    • Microsoft Fabric
    • Delta Lake
    • Python
    • Plotly
    • Power BI
    ~1,000 sites ranked
  5. 05

    Waffler - visit wafflerai.com

    A passion project I started in early 2026 as my own replacement for the likes of SuperWhisper and Wispr Flow. Hold a hotkey, talk, and it transcribes with Whisper then runs a second model pass to strip the filler and tidy the rambling before pasting clean text wherever you’re typing. Open source and bring-your-own-key, so it costs me pennies instead of a monthly subscription - and I still use it every single day.

    • Python
    • Whisper
    • Groq
    • Cerebras
    • PyInstaller
    See it

    wafflerai.com

    Visit ↗

    Free, open source, and bring your own API key - Windows and macOS.

    The Waffler website: Less typing. More waffling. Voice-to-text that works in any app.

    The app

    103k words dictated

    The desktop app: hold the hotkey anywhere, talk, and the cleaned-up text lands in whatever you’re typing into. Everything you dictate is kept in a searchable journal, with a custom vocabulary and per-provider cost tracking. That all-time counter is my own usage.

    The Waffler desktop app header, showing the journal, vocabulary and settings tabs, the push-to-talk hotkey, and an all-time count of 103,000 words dictated
    Free & open source

Where I’m useful.

AI & agents
LLM orchestration with model fallback, prompt engineering, agent workflow design, loop engineering, MCP integrations.
Microsoft & Azure
Microsoft Fabric, Azure AI Foundry, Azure OpenAI, Azure Functions, Copilot Studio, Power Platform, Microsoft Graph API.
Consulting & delivery
Requirements gathering, solution design, end-to-end build, roadmap ownership, live client demos.
Governance
Responsible, secure-by-design AI, risk & governance awareness, production-grade delivery.
Data engineering
Pipelines, data modelling and reporting layers; Python, JavaScript / Node.js, REST APIs, event-driven workflows.
Tooling
Git, Claude Code, Codex, Cursor, GitHub Copilot, N8N, the Azure stack.
  1. 01

    Scope the challenge

    Gather the requirements, understand the process.

  2. 02

    Design the solution

    Get the data in order, design the architecture - and decide whether AI even belongs in it.

  3. 03

    Build it

    Build it myself, using AI to move fast.

  4. 04

    Ship it

    Straight to production, no shortcuts.

  • Winner, Morta AI Hackathon Beat much larger firms
  • Distinction, Multiverse Data Fellowship Applied data
  • BSc (Hons) Quantity Surveying - University of Reading 2:1 · 2021
  • Digital Consultant, RLB Digital 2024 - present

Let’s connect.

Tell me the business problem - I’ll tell you whether AI is the right tool, and what it would take to ship it.

Solihull, UK