Workshop management and POS system
A Laravel application for a vehicle body repair business, covering point of sale and day-to-day workshop management.
MSc Computer Science, Distinction. University of Hertfordshire
I build software, and I measure how AI agents distort what they pass along.
Software engineer and freelance developer. I deliver working systems for real businesses, and my master's research quantified how meaning drifts and hallucinations accumulate in multi-agent language model pipelines.
Move your pointer over the signal to disturb it.
clean drifted
Research
My dissertation, Quantifying Semantic Drift and Hallucination Accumulation in Sequential Multi-Agent LLM Architectures, was awarded a Distinction.
Multi-agent systems chain language models so each one builds on the last. I built an experimental pipeline to measure how meaning shifts, and how unsupported claims creep in, as information moves from agent to agent.
- QuestionHow quickly does meaning degrade in a sequential agent chain, and can that be measured?
- ApproachA purpose-built artefact that runs controlled agent chains and scores each hand-off.
- Why it mattersAgents are being wired together in production without knowing how much each link distorts.
The same effect, in words
An illustrative example: a clear instruction loses its force across four hand-offs. Changed wording is highlighted.
- Original
The patient must stop taking the tablets and see a doctor within two days.
- Agent 1
The patient should stop the medication and visit a doctor in a couple of days.
- Agent 2
The patient may want to pause medication and consider seeing a doctor soon.
- Agent 3
Patients can consider adjusting their medication and consulting a professional.
- Agent 4
Medication changes are optional; professional advice may be helpful.
Selected work
Semantic drift test artefact
The experimental pipeline behind my dissertation. It runs sequential LLM agent chains and measures drift and hallucination at every step.
Freelance development
Client-facing software work: understanding what a business needs, building it, and handing over something they can run.
About
I'm a developer who likes systems that work for the people using them. I take projects from a client's first description to a deployed product, and I now apply the same care to understanding how AI agents behave when they work together.
I'm looking for roles where I can build reliable software and bring research habits, such as measuring first and claiming second, to the work.
- EducationMSc Computer Science, Distinction. University of Hertfordshire
- FocusMulti-agent LLM systems, evaluation, web applications
- Working asFreelance software developer