Liam Butler.

An adaptive neck support collar for people living with Motor Neurone Disease, who gradually lose the neck strength to hold their head up. It won Best Pitch at the MTU Prize for Innovation 2025.

My part: I came up with the idea of supporting the head with inflation, researched the materials, did the electronics and helped write and deliver the pitch.

• The collar is built from a ring of inflatable tubes divided into sections: front, back, left and right. Each section inflates or deflates on its own to support the head where it is needed.
• It starts with a simple self-inflating valve. As the disease progresses, a unit with an air pump takes over, so support grows with the user's needs.
• From my materials research: TPU (thermoplastic polyurethane) tubes for flexibility and puncture resistance, under a Tencel cover that is soft, moisture-wicking, hypoallergenic and resistant to bacteria, which suits users who spend long periods in one position.

• Product design
• Electronics
• Pneumatics
• Materials research
• Assistive tech
• Pitching

A redesign of the surgical trocar, the instrument used to create a port into the abdomen in keyhole surgery. Our team improved its ergonomics and safety mechanisms so insertion force can be customised to minimise patient trauma. It won Best Design at the MTU Prize for Innovation 2025.

My part: I helped with the CAD and did the electronics and software.

• Built the electronics around an Arduino Uno, which measures the force applied during insertion.
• Wrote the software that compares that force against a set limit and warns the surgeon if they are pushing too hard.
• Helped model the handle and trocar assembly in CAD.

• Arduino Uno
• C++
• Electronics
• CAD
• Medical devices
• Product design

An automatic pet feeder with remote monitoring, built in a team of three.

My part: the camera and sensors.

• Read temperature and humidity from DHT sensors and food and water levels from ultrasonic sensors.
• Integrated an ESP32 camera and sent the data over MQTT through Azure to a time-series database.
• Owners set feeding times, repeats and portion sizes from a web app.
• Validated the system with vets and pet care providers.

• ESP32
• C
• MQTT
• Azure
• Sensors

A full penetration test in two parts: OSINT reconnaissance against a large real-world organisation, then active scanning and exploitation of two vulnerable machines on an isolated VirtualBox lab network.

• Reconnaissance: passive OSINT only, using theHarvester, Sublist3r, Knockpy, WhatWeb, Recon-ng, Shodan and SpiderFoot. Mapped 180 subdomains and 112 IP addresses, mapped the findings to MITRE ATT&CK and made defensive recommendations.
• Metasploitable 2 (Linux): scanned with Nmap and Nessus, then got root through the vsftpd 2.3.4 and UnrealIRCd backdoors and recovered /etc/shadow.
• Metasploitable 3 (Windows Server 2008 R2): SYSTEM access via EternalBlue (MS17-010). Dumped 20 NTLM hashes, then chained harvested credentials into a Tomcat Manager WAR upload for a second SYSTEM shell.
• Wrote prioritised remediation for each target: patching, least privilege, MFA and secrets management.

• Kali Linux
• Nmap
• Nessus
• Metasploit
• Shodan
• SpiderFoot
• MITRE ATT&CK

A web app that tackles "what should I read next?" by combining a user's own star ratings with book metadata. We built it in Scrum sprints, and I was Scrum Master for Sprint 1.

• NiceGUI front end with registration and login. Passwords are stored as salted PBKDF2 hashes.
• Over 11,000 books in MongoDB Atlas, searchable by title or author. Genres come from the Open Library API and descriptions from Google Books, both looked up by ISBN.
• I built the rating feature: star ratings with one rating per user per book, which can be edited later. Users also have a wishlist and recommendations based on the books they rated highest.
• Ran the project in Jira (user stories, story points, planning poker, burndown charts), with feature branches and pull requests on GitHub.

• Python
• NiceGUI
• MongoDB
• REST APIs
• Jira
• Scrum
• Git

Analysed mobile app usage data to find out which users the app keeps and which it loses.

• Segmented users with K-Means and DBSCAN into three groups: Retained, Needs Attention and Churn.
• Balanced a heavily skewed dataset with SMOTE, then trained logistic regression and KNN classifiers that reached 100% recall on users who uninstalled.
• Wrote a report with recommendations for targeted retention campaigns.

• Python
• scikit-learn
• K-Means
• SMOTE
• UMAP

A series of IoT labs that collect, process and sync data between Raspberry Pis at the edge and the cloud.

• Sensor monitoring: IOTstack (Docker) on two Pis, with DHT11 temperature and humidity readings going over MQTT into PostgreSQL and InfluxDB, and Node-RED, Flask and Grafana on top. Fast temperature alarms (>50 °C within 100 ms) run at the edge, and sustained-humidity warnings run in the cloud.
• InfluxDB: Telegraf logs Pi health data (CPU temperature, Mosquitto and Node-RED status) to a local InfluxDB, which replicates it to Azure InfluxDB SaaS.
• PostgreSQL: logical replication of a users table from a cloud database to the Pi.
• Final project: two NiceGUI single-page apps. One is a user management app (insert, update, soft and hard delete) hosted on Azure Container Apps. The other is an offline login app on the Pi that launches in Chromium kiosk mode at boot through systemd.

• Raspberry Pi
• Docker
• MQTT
• Node-RED
• Grafana
• PostgreSQL
• InfluxDB
• NiceGUI
• Azure

Notebooks analysing HVAC IoT sensor data loaded from PostgreSQL. I started with exploratory analysis, then built linear and logistic regression, tree-based models and an MLP neural network to predict fault state, energy use and comfort.

• Python
• pandas
• scikit-learn
• PostgreSQL
• Jupyter

A team of three built an FPGA hardware platform for video games on the Basys3 board, written in Verilog. It has 640×480 VGA output, keypad and joystick input, settings saved to serial NOR flash, and a clean game-module interface so new games can plug in without hardware changes.

• I owned the input subsystem: a custom SPI master for the Pmod JSTK2 joystick and a scanning FSM for the 16-button Pmod KYPD keypad.
• Wrapped both behind one interface that gives the rest of the system stable, debounced and synchronised signals.

• Verilog
• Vivado
• Basys3
• SPI
• VGA