Systems I have built, tested and delivered

Six AI and automation systems I built and tested myself, and five enterprise projects where I worked inside a larger delivery team.

AI and automation systems

For each one, the problem it solved, what I built and how I checked that it works.

Diagram of the automation: an Excel input sheet, validation, terminal automation, a state check, a run ledger and a safe stop before the next client

Insurance Workflow automation

Linux Reflections Workflow Automation

Problem: A repetitive PAS anniversary workflow required hours of manual terminal navigation and data entry, with interruption creating duplicate, missed-work and recovery risk.
My solution: Built an Excel VBA automation using Reflection for screen observation and Windows key injection for actions, with controlled inputs, logging and rerun logic.
Evidence: Validated compile, navigation, single-client, round-trip and batch behaviour in the live system; rerun testing correctly skipped 83 clients already recorded as successful.

AI-Assisted Engineering
Quality Gates
Guardrails
Human-in-the-Loop

Diagram of the pipeline: Microsoft Teams capture, transcript retrieval and clean-up, then AI processing into a summary, decisions and actions by person, with ambiguity flags and human review

OIC Architects AI automation

Meeting Intelligence Pipeline

Problem: Teams transcripts captured important decisions and actions, but left them buried in recordings and unusable by the workflows that followed.
My solution: Built the transcript capture and processing pipeline, with structured notes, routing rules, ambiguity flags, human review and downstream action handling.
Evidence: Tested repeat processing and compared independent AI outputs, exposing duplicate transcripts and under-specified formatting rules. Fixes went into controls and specifications rather than blind model tuning.

Power Automate
Human-in-the-Loop
Guardrails
Failure Modes
Context Engineering

Diagram of how query quality is checked: interpret the business intent, construct the joins and conditions, then validate the generated SQL against the intended result

Insurance Test data

Natural Language SQL Query Engine

Problem: Complex test-data queries often required five or six tables; unfamiliar joins could fail or create hard-to-explain duplicate results.
My solution: Built a Copilot-based natural-language SQL engine, onboarding table relationships, field meanings and business rules around policies, lives and benefits.
Evidence: Used it on real test-data searches, checking generated queries against the intended conditions and distinguishing genuinely missing data from query-construction problems.

GitHub Copilot
Context Engineering
NL-to-SQL
Software Quality

Diagram of the exam app: an AI pipeline that generates and scores practice questions against a rubric, with quality gates before each rebuild

Exam preparation AI evaluation

Architect Registration Exam Preparation App

Problem: Existing practice banks could be gamed through answer-position and option-length patterns, weakening the score as a measure of actual knowledge.
My solution: Defined the rubric and quality gates, built the agent pipeline that generates and scores questions, and shipped the offline practice app.
Evidence: Calibrated the rubric against 37 exam-board sample questions and gated rebuilds on 12 numeric checks, catching 89.7% answer-position bias and systematic option-length tells.

Offline Evals
Benchmark Calibration
Deterministic Graders
Regression Evals

Diagram of the research pipeline: channels collected, scored against a framework, checked by validators and turned into a structured decision brief

Competitor research AI automation

Automated Content Research System

Problem: Weekly competitor research was manual and inconsistent, producing large volumes of information without reliably identifying the strongest topic to pursue.
My solution: Built the eleven-channel research pipeline, scoring model, data contract and validator that turn competitor activity into structured decision briefs.
Evidence: Evaluated a delivered report against the original requirements and converted every identified gap into permanent rules or validators, with each validation rule tested independently.

AI Evaluation
Grounding
Risk-Based Guardrails
Human Verification

Diagram of the release gate: automated checks and Playwright visual regression at four screen widths before a website change is published

Website Test automation

AI-Assisted Website Quality System

Problem: A 39-page website was being built rapidly with AI-generated code, where a small change could silently break unrelated pages or layouts.
My solution: Directed the AI-assisted build and defined what "correct" meant, turning design and quality requirements into automated checks and release gates.
Evidence: A 52-check pre-publish gate runs alongside Playwright visual regression at four viewports. Intended changes pass, and unexpected movement anywhere else fails the gate.

Test Automation
Visual Regression
Quality Gates
AI-Assisted Delivery

Enterprise delivery

Projects where I worked as a business analyst, tester or test manager, with the context and my part in it.

Automotive

Volkswagen Car Configurator

Context: Volkswagen was rebuilding its online car configurator within a wider digital programme spanning the customer configurator, dealer contact centre and CMS.
My role: As Business Consultant, I gathered and maintained requirements, wrote user stories, worked with QA on review and testing, supported defect triage, prepared UAT scenarios and reported to the Product Manager.

Business Analysis
Requirements Engineering
QA Collaboration
Defect Triage
UAT

Life insurance

TAL Homeowner Protection

Context: A new protection product required changes to TAL's policy administration environment and close alignment between product rules, development and testing.
My role: As Business Analyst for policy administration, I reviewed specifications, clarified requirements, wrote and maintained user stories, answered delivery questions and supported business UAT.

Business Analysis
Requirements Engineering
Business Rules
User Stories
UAT

Life insurance

SailPoint Identity & Access

Context: TAL introduced SailPoint to manage identity and access for a policy administration system used by multiple business teams with different permissions.
My role: I mapped how teams used the system, translated those needs into 100+ access profiles, configured and tested them, and coordinated business UAT.

Access Modelling
Business Analysis
System Configuration
Testing
UAT

Life insurance

Fineos Claims Migration

Context: Claims functionality was migrated to Fineos while staying connected to TAL's policy administration environment.
My role: I focused on API, system and integration testing, including test data setup and verification that claims and status changes moved correctly across systems.

API Testing
Integration Testing
Migration Testing
Test Data
Cross-System Validation

Connected audio

Teufel / Raumfeld Global Testing

Context: Connected audio products were tested across six countries using real combinations of devices, apps, ISPs, routers and user environments.
My role: As Test Manager at Applause, I managed the distributed test programme, tester participation, execution, defect triage and client reporting across thousands of test cases.

Test Management
Distributed Testing
Connected Devices
Defect Management
Client Delivery

The full history, role by role, is in my CV.

Read my CV Get in touch