Applied AI · Belfast, Northern Ireland

Applied AI systems for messy real-world workflows.

I build AI products that live inside existing business processes — discovery, workflow design, model evaluation, backend implementation, integration, and iteration with real users. Strongest where the problem is ambiguous and the solution has to fit the business, not the demo.

View work Causeway case study CV

Selected work

Causeway

AI course generation · EdTech · Brownfield integration

An education platform used by seven medical schools internationally. Sales were hard because prospects couldn't quickly see their own material transformed — so I worked backwards from that bottleneck to an AI generation layer: educators upload existing content and the system turns it into interactive course material.

Single-element generation live with paying institutional users
OpenAI, Claude, and Gemini evaluated against the real task — not benchmarks
Python/FastAPI AI service integrated into a brownfield Laravel product
Document ingestion, chunking, embeddings, structured generation, Logfire observability
CRM lifecycle hooks and Stripe commercial flows

Read the case study →

Elvante

Forward-deployed AI enablement · Governance · Apprenticeship delivery

Implementation tutor on a Level 4 AI & Automation apprenticeship — dropped into client organisations with almost no handover and expected to find the workflow, set the data boundary, and deploy AI usefully on two hours a month per client.

Two client organisations live — funded-learning compliance and a software agency — two more onboarding
GDPR-sensitive workflows automated without data leaving the client environment
Live mid-session pivot from prompt patterns to integrated agent tooling on the client's real ticket queue
Governance turned into deliverable structure: employer approval gates, evidence trails, auditable session records

Read the case study →

Local RAG Workbench

Local-first RAG · Medico-legal · Evaluation

A source-first evidence workbench for expert witnesses reviewing large, confidential medical-legal bundles. Not a PDF chatbot: a review desk for finding, verifying, and organising facts — where nothing leaves the machine.

Fully local: Ollama models, Docling with OCR/VLM extraction
Hybrid retrieval — ChromaDB, BM25, reranking — with source-first citations
Public eval suite scoring page-hit and required-term coverage against a synthetic bundle
In use with a real client; the eval results drive the roadmap

Code and evals on GitHub →

Prospect Intelligence

Data pipeline · CRM automation · Human-in-the-loop

A vertical prospect-intelligence engine for Flow Local's outbound operations. It turns source-of-truth directories and business signals into scored, reviewable, CRM-ready outreach lists — in daily use, producing replies.

Source scraping, domain discovery, enrichment, and scoring per vertical
Human QA stage where automation is imperfect: weak domains, bad emails, poor fit
Notion review flow activating into the GoHighLevel CRM

Code on GitHub →

How I work

DiscoveryWork backwards from the commercial or operational pain, not from the model.
Workflow designMap how the work actually happens — constraints, handoffs, and data boundaries included.
Model evaluationTest providers against the real task with real evals. Pick on fit, not preference.
Build & integrateShip inside the existing product — brownfield code, CRM, billing, observability.
Iterate with usersEvals, telemetry, and user feedback decide what happens next.

About

Before engineering, I spent seven years as a legal negotiator in fintech — I learned workflows, constraints, and stakes before I learned Python.

That background shapes how I build: start with the process and the people in it, treat confidentiality and data boundaries as design inputs, and measure whether the thing actually helps.

Now a product engineer in Belfast building AI systems for education, legal, and local-service businesses — from RAG pipelines and LLM evaluation to CRM integration and the unglamorous glue that makes deployments stick.

Contact

Working on something with a messy workflow at the centre? Get in touch.

michael@mcglademedia.com github.com/mikemcg87 Download CV