Open to ML / Applied AI Engineering roles

Building intelligent systems
at the intersection of
AI & security.

// AI and Cybersecurity Researcher

3rd-year Artificial Intelligence & Computer Science undergraduate at the University of Edinburgh, building ML systems that work on real data and solve real problems.

I specialise in converting complex, real-world data streams into clean, structured data through the development of modular pipelines.

My work ranges from video-analysis systems for judo, to anonymization tools for email corpora.

I recently completed an AI & Cybersecurity research internship at the University of Edinburgh's School of Informatics.

View my work Get in touch

featured projects

paper → ACM '26

AI+Sec

research intern

01 / about

Turning research into real systems

Hi, I'm Conor! I'm an AI and Cybersecurity Researcher with strong engineering skills.

I enjoy taking messy, unstructured data streams and converting them into clean, usable formats for AI models, then communicating those solutions clearly to others.

Outside of work, I train Judo, which led naturally to my most recent project: an AI judo coach for post-match analysis. Right now it takes a raw match video and automatically clips out every attempted throw. The next stage is training and deploying a model to provide coaching insights, breaking down why each attempt succeeded or failed.

Toolbox

languages

PythonJavaJavaScript

ml / ai

PyTorchUltralytics YOLO v11LSTM / RNNsLLM evaluationNER evaluation

security & data

PresidioData preprocessingAnonymization

tooling

VercelKaggleGit

02 / research

AI & Cybersecurity Internship

Recently completed research examining how robust modern language models are when facing deliberately obfuscated malicious content.

Paper · Submitted to ACM 2026

An Analysis of Factors Affecting LLM-based Phishing Detection

As a contributor to this research, I manually reviewed 655 emails to curate a balanced 160-email dataset for the study, 80 using HTML/CSS obfuscation techniques and 80 without.

Curated a balanced 160-email evaluation dataset (80 obfuscated / 80 clean) from a manual review of 655 emails.
Built a Streamlit labelling interface that cut per-email labelling time by ~4×.
Reviewed and gave feedback on the final paper write-up.

Role: ContributorFocus: LLM security · Phishing detectionVenue: ACM 2026 (submitted)

03 / projects

Selected work

End-to-end builds, from model research to deployed applications and teaching resources.

Judo Clipper

PyTorchPythonJavaScriptUltralytics YOLO v11Vercel

AI-powered system using YOLO pose detection as input to a custom-built LSTM model, automatically extracting every throw-attempt clip from a full judo match.

System repo

Research repo Write-up

Email Anonymization Tool

PythonPresidio

A pipeline for anonymizing email corpora for the purpose of protecting the privacy of victims within open-source email datasets, so they can be used safely for research.

Pipeline repo

Write-up

EdinburghAI Workshops

PythonPyTorchKaggle

Self-authored resources for AI-focused workshops delivered to students at the University of Edinburgh's student-led AI Society.

concepts taught

YOLO Pose DetectionData PreprocessingLSTMRecurrent Neural Networks

Resources repo

Society contributions Write-up

04 / contact

Let's build something.

I'm actively looking for ML / Applied AI Engineering opportunities and open to research collaborations. Feel free to reach out.