Sebastian SouthwellMultidisciplinary Software, Systems & DevOps EngineerCloud · Software · AI · Hardware
Who am I?
I'm Sebastian Southwell, a multidisciplinary software and systems engineer with a First-Class MComp in Computer Science from the University of Sheffield. Currently a DevOps Engineer at the UK Home Office, I build reliable, scalable systems and am particularly interested in platform engineering, applied AI and proof-of-concept development. My academic work includes leading master's-level research into reducing hallucinations in large language models and shaping the design and strategy of a start-up development project.
Outside work, I've contributed to a wide range of projects and technical communities. I won prizes at HackSheffield 7, 8 and 9 and held several university leadership positions, including Technical Systems Officer, Technical Events Officer, Inclusions Officer and Lead Imaging Systems Engineer for a student-led UAV project.
I'm also active in the Sheffield community, have been partner dancing for two years and competed at IVDC Blackpool in 2025.
Leading migration of the team's development environment between Cloud IDEs — experimenting with and implementing a solution alongside a cloud platform representative. Onboarding various new applications into our primary infrastructure, configuring them and proving value while working closely with stakeholders.
Associate DevOps EngineerFeb 2024 - Dec 2025
Implemented a PrivateLink for high-traffic network infrastructure, reducing unnecessary data transit. Built dynamically reproducible, interconnected networks with Terraform and Packer at the request of an internal team, ensuring fit for requirements and developing a proof of concept. Created permission-auditing scripts to help enforce a zero-trust policy across our services and team resources. Sat on interview panels and mentored junior engineers.
Apprentice DeveloperSept 2022 - Feb 2024
Working with a senior engineer, built an autoscaling data forwarder handling 2,000+ events per second. Produced a VPN connection automation script that reduced on-device connection and setup time for engineers. Contributed to a government backend project modernising certificates and records, working with QA and following quality review processes.
The Catalyst
Marketing OfficerNov 2022 - Jan 2024
Built a front-end website in Jekyll on GitHub Pages for dynamic display and easy modification, and wrote documentation in LaTeX. Performed and tracked performance of social media posts, and also initiating automatic stat monitoring using LinkedIn and Instagram API.
Societies Committee (Sheffield Students' Union)
Technical Systems OfficerAug 2022 - May 2023
Built digital solutions to increase interactivity between the societies committee and the public via tools like Apps Script, automating publicity such as emails via scripting.
Used Canva for announcements and HitFilm Express for an animated awards short. General committee member voting on funding and other matters.
Created an automatic bin-packing inspired script that would allow for the efficient placement of many individuals on the DYSO University barcrawl of which there were 2000 attendees with various societies, requiring optimal route placement both for walking distance and the bar capacities.
Media & Technology Societies LiaisonFeb 2022 - Aug 2022
Liaison for media and tech societies and general committee member; assisted with setup and announcing at the activities awards. Attended regular meetings and voted on funding allocations.
Spoke on stage and presented awards while also acting as stage wrangler, assisting with the setup and management of the university's annual society awards ceremony.
Sheffield Ethical Student Hackers
SecretaryMay 2022 - Nov 2022
Attended regular meetings, assisted with presentations and assisting existing committee members, while also actively taking minutes in meetings, and controlling room bookings.
Inclusions OfficerSept 2021 - 2022
Co-delivered weekly cybersecurity sessions to groups of around 10 students, teaching responsible use alongside both offensive and defensive techniques including web scraping, SQL injection and ethical hacking.
Created challenges and lesson plans for sessions, and ensured they were accessible across all skill levels, alongside a general committee role covering session planning and website development with git. Worked with AWS, maintained their professional website and performed updates, while also writing capture the flag challenges for our annual competition.
Project Hex
Lead Imaging Systems EngineerMar 2022 - Nov 2022
Managed and instructed a team of 4-5 different engineers, working on a drone project for the US SUAS competition. Attended regular management meetings, and gave timeline and Gantt chart updates while also acting as a panel member for interviews, assessing technical ability of candidates.
Imaging Systems EngineerOct 2021 - Mar 2022
Student-led engineering project competing in the US SUAS competition. Handled and formatted data sets in an automated fashion for compatibility, and researched TensorFlow models for use on a UAV with an onboard Jetson Nano.
Sheffield CompSoc
Technical Events Officer2021 - 2022
Designed and delivered a Python for Beginners workshop and a C++ for Beginners workshop to undergraduate audiences, with the C++ workshop being an original initiative. Organised two Year in Industry workshops, coordinating speakers and logistics, and assisted students hands-on throughout sessions, adapting to last-minute changes and absences.
Judged, assisted, planned and managed the University of Sheffield ShefJam 8 competition, a 42 hour game development competition and contributed to the round-the-clock safety and event management of the competition working in shifts.
Academic Experience
The University of Sheffield
Master of Computing (MComp), Computer ScienceSept 2020 - Jun 2025
First-Class MComp graduate whose degree centred on three major projects:
Darwin (BDUK) — System Architecture lead on a graph-verified RAG assistant mitigating hallucinations in multi-modal LLMs, built for government body Building Digital UK (74%).
Dissertation — a physical smart-home voice assistant on the University of Sheffield's unPhone hardware platform, combining LLMs, speech processing and IoT, supervised by Professor Hamish Cunningham (78%).
Genesys — a Ruby on Rails career-guidance platform for University of Sheffield spinout epiGenesys, covering the full product lifecycle from customer interviews, personas and pain-point/experience mapping through to a business model and the build itself — praised for its testing, security and performance engineering (75%).
Alongside the technical curriculum, Finance and Law for Engineers (88%) covered financial statement literacy (income statement, balance sheet, cash flow), UK company law and audit regulation, and pricing strategy grounded in demand economics — the commercial and legal literacy behind the engineering.
First-Class MComp graduate with a strong AI focus spanning real projects, practicals and theory:
Darwin (BDUK) — System Architecture lead on "Mitigating Hallucinations in Multi-Modal LLMs: A Graph-Verified Retrieval Assistant for Government Data," combining dual embedding models (local CLIP, cloud-based Amazon Titan) and two document-verification approaches, benchmarked across 275 configurations (74%).
Dissertation — a physical smart-home voice assistant on the unPhone hardware platform, combining LLM-driven command generation, speech processing (recognition and synthesis, with wake-word detection researched but not reliably implemented), and a Flask/Redis cloud backend, supervised by Professor Hamish Cunningham (78%).
Speech Processing (80%) — built custom digital filters and a suite of DSP-based voice effects from scratch in Python (tremolo, ring modulation, frequency shifting, vibrato, echo, comb filtering, a flanger, and a cascaded equaliser).
Speech Technology (73%) — chose keyword search (information retrieval over spoken audio) and a review of the DIHARD speaker diarisation challenges as practicals, from a curriculum covering MFCCs, HMMs and other graphical models, Gaussian Mixture Models, and n-gram language models.
Machines and Intelligence (72%) — my first AI module: implemented A* and Branch and Bound search in Java for "The Rambler's Problem," then wrote a batch-automation harness plus a separate C# analysis tool to run around 6,885 systematic test variants comparing how heuristic weighting affected search efficiency.
Text Processing (63%) — built a TF-IDF-based document retrieval system, alongside labs in Chinese word segmentation and part-of-speech tagging.
Data Driven Computing — a full pattern-recognition and machine learning curriculum studied in depth: Bayesian decision theory, linear classifiers and the Perceptron, k-nearest-neighbour classification, PCA/LDA-based dimensionality reduction via eigenvectors, and deep learning through to CNNs and RNNs.
First-Class MComp graduate with core software engineering training:
Software Reengineering (73%) — coupling/cohesion metrics, the SOLID object-oriented design principles, refactoring, code-clone detection, and legacy-system reengineering patterns.
Software Hut — a team-built real software project for an external client (86%).
Genesys — a Ruby on Rails career-guidance platform for epiGenesys, spanning customer interviews and business-model work through to a build marked highly for its testing, security and performance engineering (75%).
Finance and Law for Engineers (88%).
First-Class MComp graduate whose degree included a Year in Industry placement — the basis of my current DevOps career:
The Internet of Things — built network-connected devices on ESP32 and the unPhone hardware platform, including sensor/actuator integration and MQTT-based cloud messaging (96%).
Cyber Security Team Project (72%) — designed security controls and network segmentation for realistic business scenarios, ran Cyber Essentials-style assessments for client organisations, and completed threat-modelling and intrusion-detection exercises.
Genesys — contributed CI security checks, dependency audits, and automated testing to a production-style Ruby on Rails deployment (75%).
First-Class MComp graduate with full-stack coverage across the degree:
Web and Internet Technology (86%) and Software Hut (86%).
Genesys — a full-stack Ruby on Rails career-guidance platform for epiGenesys, marked highly across coding, testing, security and performance (75%).
Dissertation — combined a Flask/Redis cloud backend with a physical unPhone hardware front end (78%).
First-Class MComp graduate with front-end-relevant coursework:
Web and Internet Technology (86%).
3D Computer Graphics (74%) — an assignment comparing Blinn-Phong against Physically Based Rendering for real-time photorealistic floor rendering across marble, wood and carpet materials, covering Lambert's Law, microfacet distribution and the Torrance-Sparrow BRDF.
Genesys — front-end work on a Ruby on Rails platform for epiGenesys, recognised for its accessible, keyboard-navigable UI (75%).
First-Class MComp graduate with back-end-relevant coursework:
Genesys — back-end development on a Ruby on Rails platform praised for eager loading, scoped ActiveRecord queries, and service-object architecture (75%).
Dissertation — a Flask/Redis cloud backend tied together via a RESTful API (78%).
First-Class MComp graduate with hardware-adjacent coursework:
The Internet of Things — built network-connected devices on ESP32 and the unPhone hardware platform (96%).
Devices and Networks (92%).
Robotics (65%) — programmed a simulated mobile robot in ROS/Gazebo, working with custom message, service and action definitions for sensing and control.
Dissertation — a physical voice assistant built on the unPhone hardware platform (78%).
AWS Certified Cloud PractitionerFoundational certification covering core AWS services, cloud architecture and security best practices.CLF-C022024 – 2027
Extracurricular
Partnered Dance
Latin & Ballroom, Salsa and Bachata.
Competed in Latin & Ballroom at Blackpool, plus social dancing across all three styles: footwork, frame, connection and musicality. Regularly attending community events.
Technical Reels
Making short technical/coding reels on Instagram. Quick explainers and builds in bite-sized video form, alongside occasional travel vlogs. @seb.not.found
Improv Comedy Jams
Attending regular improv comedy jams, of which the initial objective was to improve public speaking, quick-thinking improvisation and build character.
Get in touch
Message sent!
Something went wrong — please try again in a moment.
Genesys — Career Guidance Platform for epiGenesys
A two-semester, master’s-level team project for COM4525, completed by a six-person team in partnership with the University of Sheffield spinout epiGenesys. The goal was to develop a startup minimum viable product using Lean Startup practices.
I proposed the initial concept: a career-guidance platform that helps users explore possible career transitions based on transferable skills rather than job titles. Instead of presenting careers as a traditional ladder, the system visualised them as an interconnected tree, showing how users could move between different career branches from their current position.
I led much of the initial discovery and product research, including 30 to 35 customer interviews, persona development, user-experience and pain-point mapping, story cards, and the creation of a Lean Canvas business model. These findings shaped the final product brief and ensured that development was grounded in validated user needs.
On the engineering side, I worked across the Rails 7 stack and led the implementation of the JavaScript-based career graph. The matching engine used breadth-first search to calculate hop distances between careers and scored potential transitions against the SFIA skills framework. A Delayed Job background worker allowed the graph to be recalculated without blocking user requests.
I maintained oversight of the project’s most technically complex features, with particular responsibility for testing, performance, security, and accessibility. Security was implemented using Devise authentication and CanCanCan role-based access control, alongside Have I Been Pwned k-anonymity password checks. The application was also assessed using Brakeman and Bundler Audit static security scans.
Testing included RSpec unit tests, Capybara and Selenium end-to-end workflows, performance testing, and an internally developed analytics layer for monitoring lead capture and conversion without relying on third-party tracking. I designed the structure of the team’s testing documentation and directed its production, which received the project’s highest individual assessment area.
A dedicated user-experience testing phase included moderated think-aloud sessions, WCAG contrast checks, keyboard-only navigation, and simulated colour-vision constraints. This process identified genuine accessibility issues, including low-contrast controls and filter dropdowns that could not be accessed using the keyboard. These issues were corrected and then retested, with libraries selected specifically to support reliable tab navigation.
The project received a mark of 75% and was recognised particularly strongly for its testing documentation, security, performance engineering, and accessibility work.
Client:
epiGenesys (University of Sheffield spinout)
Date:
Oct 2024 - May 2025
Service:
University
Dissertation — A Physical Smart-Home Voice Assistant
My final-year dissertation, supervised by Professor Hamish Cunningham, focused on the development of a physical smart-home voice assistant using the University of Sheffield’s unPhone hardware platform. The device was ESP32-based and included a touchscreen and onboard microphone.
The accompanying design figure is my own Fritzing-based proposal, showing how the device could be connected to a breadboard and how its on-screen feedback states might appear. On the hardware side, audio detection triggered recording and streaming to the backend for processing.
The system used a combination of large language model APIs and locally hosted models to interpret commands and evaluate generated outputs. I proposed several evaluation methods, including BERTScore and a range of custom metrics. This work was completed before structured LLM output schemas were widely available, so the generated responses were considerably more unpredictable and experimental.
As part of the project, I researched a broad range of hardware, software, text-processing, and audio-processing techniques. This included investigating robust wake-word detection, although the final wake-word implementation was removed from the project scope due to time constraints.
On the backend, a Flask and Redis service exposed through a RESTful API handled speech recognition and passed transcribed commands to an LLM for interpretation. I also designed and built a dynamic smart-home environment using Redis and additional microcontrollers, allowing the system to demonstrate an adaptable and intuitive smart-home experience.
The project was recognised as particularly ambitious for a third-year dissertation and ultimately achieved a first-class mark of 78%.
Client:
University of Sheffield (Dissertation)
Date:
Oct 2024 - May 2025
Service:
University
Darwin — Graph-Verified RAG Assistant for BDUK
A two-semester, master’s-level group research project for COM4520, completed in collaboration with the government body Building Digital UK. The project investigated methods for reducing hallucinations in large language models and enabling government data to be used in a reliable, secure, and informative way for employees.
Due to unforeseen circumstances, the team was without an academic supervisor during the first semester. I therefore assumed a leading role within the group, helping to define the project scope, establish the research direction, and coordinate the investigation. I led the system architecture and focused my research on agentic approaches, retrieval techniques, and output-verification strategies.
We evaluated a wide range of variables, including embedding models, language models, retrieval methods, preprocessing strategies, and verification techniques. As part of this work, we devised a custom verification method inspired by knowledge graphs and logical axioms. The approach aimed to decompose source documents into objective statements that could be compared against generated answers, providing a structured method for identifying unsupported claims.
The final system consisted of separate preprocessing and query-time pipelines. During preprocessing, documents were uploaded to Amazon S3, divided into multimodal elements, and prepared for retrieval. Infrastructure was provisioned using Terraform and AWS, with secure, least-privilege access to Amazon Bedrock. Preprocessing reduced query latency and token consumption by completing computationally intensive work before user requests were received.
Content was embedded using two approaches. Local CLIP embeddings combined textual and visual information within ChromaDB, while Amazon Bedrock Titan generated cloud-based multimodal embeddings. Video and audio content could also be processed, transcribed, and incorporated into the retrieval system. In parallel, an LLM-based graph transformer extracted entities and relationships into a Neo4j knowledge graph.
At query time, the system combined vector similarity search with structured graph traversal. This hybrid retrieval approach allowed generated answers to be checked against explicit entity and relationship evidence rather than relying on semantic similarity alone. Verification was further supported through subject, action, and verb parsing to compare responses against information extracted from the knowledge graph.
The architecture was designed to remain provider-agnostic. A configurable model catalogue supported OpenAI models, Amazon Bedrock models such as Claude, Llama 3, and Mistral, as well as Gemini, xAI, and locally hosted Ollama models including DeepSeek R1, LLaVA, and Qwen2.5.
We assessed each configuration using BERTScore and ROUGE-L across 275 benchmark runs. The project also included a pytest test suite, GitHub Actions continuous integration and deployment, Discord build notifications, and secure credential management throughout.
To demonstrate the system, we developed an interactive Streamlit interface that allowed users to submit queries, explore retrieved evidence, and review generated responses. As a team, we also communicated our research through interim presentations and a final academic poster presentation, explaining the system architecture, experimental methodology, findings, and practical implications to both technical and non-technical audiences.
The project received a mark of 74%.
Client:
Building Digital UK (BDUK)
Date:
Oct 2024 - May 2025
Service:
University
Property Management App for Regent Court Locksmiths
A real-client Android app (COM4510 team project) built for Regent Court Locksmiths, letting landlords manage properties and raise repair requests with their contractors. Built in Kotlin with Jetpack Compose, it supports full property/repair CRUD with photo attachments, an embedded Google Maps picker (with a custom pointer-gesture fix to stop the map fighting with its own bottom-sheet modal), auto-composed repair request emails with attachments, and phone number auto-detection via Android's TELEPHONY_SUBSCRIPTION_SERVICE. Requirements were gathered through customer personas and QOC (Quality-Options-Criteria) decision tables to weigh UX trade-offs as a team of four. Delivered using a proper agile workflow: Trello Kanban, git feature branches with peer-reviewed merge requests, and a GitLab CI/CD pipeline wired into Discord for build notifications.
Client:
Regent Court Locksmiths
Date:
Nov 2024 - Dec 2024
Service:
University
Eye Of Horus — AI-Rendered Open-World Simulation
Inspired by Oasis AI's real-time AI-rendered Minecraft, this 24-hour hackathon build asked whether the same idea could work for a physical device instead of a screen: a Kitten-box handheld wired to a joystick, displaying an AI-generated "open world" that responds to navigation input. We wrote the analog joystick and image-display firmware directly on KittenCode online — working around limited firmware control and no native image support, which meant designing a custom image format and a manual drawing approach — while Databricks, using OpenAI's DALL-E and LangChain, generated the environment images from user prompts behind a Streamlit frontend. Like all our hackathon builds, the strategy in a 24-hour competition was to deliberately reach for as many different technologies as possible, treating the time pressure as a fast learning opportunity rather than just a race to ship. Won [MLH] Best AI Project with Databricks Open Source at hackSheffield 9.
Client:
hackSheffield 9
Date:
Nov 2024
Service:
Hackathon
3D Computer Graphics — Assignment 1: A Custom Scene
A from-scratch 3D scene developed in Java with JOGL, the Java bindings for OpenGL, for COM4503. Rather than relying on a ready-made framework, I designed and implemented my own hierarchical scene-graph engine for managing geometry, transformations, animation, lighting, and rendering.
The architecture was designed around dynamically controllable scene components and swappable meshes. A base SGNode class handled recursive tree traversal through draw, update, and disposal operations using a composite-pattern structure. A TransformNode subclass stored transformation matrices, while ModelNode leaf objects connected the scene graph to renderable geometry. This separation allowed meshes, materials, transformations, and behaviours to be replaced or controlled independently without tightly coupling scene structure to the rendering implementation.
The project involved substantial use of 3D mathematics, including vectors, matrices, coordinate-space transformations, hierarchical world transforms, Euclidean-distance calculations, rotations, scaling, translation, and animation interpolation. Parent and child transformations were composed throughout the graph so that complex objects could be moved and animated while preserving the relative positions of their individual parts.
For multi-part objects such as the robots, I created a reusable foundationTransforms() factory method that automatically generated scale, translation, and rotation node chains. This avoided repeating transformation boilerplate for every limb and made articulated models easier to construct, modify, and animate.
I also devised and implemented a complete keyframe animation system. This allowed object transformations to be interpolated between defined animation states, supporting coordinated movement across multiple scene-graph nodes. Robot1’s dancing behaviour used this system and was triggered through a Euclidean-distance method that I added to SGNode. The method compared world-space positions to detect when Robot1 moved close to Robot2, which independently patrolled the scene.
The rendering pipeline used custom shaders that I wrote myself, with Blinn-Phong-based lighting calculations for ambient, diffuse, and specular illumination. The scene included both a point light and a spotlight, requiring lighting and surface calculations to be performed within the shader pipeline.
Additional features included a spacecraft controlled using WASD and the arrow keys, a skybox that followed the camera, and a textured globe rotating on an axle. All textures were created from scratch in GIMP, including the repeating dice-logo wall texture, which uses the same personal logo featured in this website’s radial menu.
The screenshot was captured by recompiling and running my original submission. This assignment was assessed alongside Assignment 2, with both assignments collectively contributing to a module mark of 74%.
Client:
University of Sheffield
Date:
Oct 2024 - Nov 2024
Service:
University
IoT Robot Car — Live Streaming, IR Control & OTA Updates
A COM3505 Internet of Things assignment involving the design and construction of a remotely controlled robot car using a DFRobot FireBeetle ESP32 development board. The system combined live video streaming, infrared command reception, and motor-shield-based control of speed and rotation.
The ESP32 hosted a dedicated web interface with separate pages for basic controls, advanced controls, Wi-Fi configuration, and over-the-air firmware updates. Through this interface, users could view the live camera feed and manually control the car’s movement directly from a browser.
Firmware updates were delivered using ElegantOTA, allowing new versions to be installed wirelessly without repeatedly reconnecting and reflashing the board over USB. Each hardware component was deliberately selected according to the project’s functional and performance requirements rather than relying only on components that were readily available.
Alongside the car, I independently designed and built a standalone infrared remote controller. The controller used a joystick connected to an infrared emitter to transmit movement commands to the vehicle. Raw joystick coordinates were converted from Cartesian x and y values into polar coordinates, using the resulting angle and magnitude to determine the car’s direction and speed.
A group project for COM3523 — my own individual contribution was a Python tool that walks the real commit history of the bottle web framework — checking out every historical commit that touched its main source file, and at each one running static analysis: a UML class diagram via pyreverse and Graphviz (shown here), a call graph, and a lines-of-code count exported to CSV. Stitching every commit's diagram together produces a short animation of the codebase's structure evolving commit-by-commit across its real development history — a practical look at coupling, cohesion and how a real open-source project's architecture grew over time, tying directly into the module's coverage of the SOLID principles, refactoring and legacy-system reengineering patterns (73%).
Client:
University of Sheffield
Date:
Jan 2024 - Apr 2024
Service:
University
Redcommer — Reddit Comment Merger and Visualizer
A 24-hour hackathon build that turns Reddit comment history into AI-generated imagery: pick a handful of subreddits, and a sampled set of their 2011-era comments (pre-processed down from ~15 million to ~2 million) is pulled from a MongoDB Atlas store, summarised and reformatted into an image prompt by ChatGPT, then rendered with DALL-E. I worked on the backend data server — the real engine behind the Streamlit frontend — wiring up and debugging three separate third-party APIs with Postman, each needing its own handler. As with our other hackathon entries, the strategy going in was to deliberately combine as many different technologies as possible in the 24 hours, using the time pressure as a learning opportunity rather than just a race to ship. Won Best Hack 3rd place and MLH Best Use of MongoDB Atlas at HackSheffield 8.
Client:
HackSheffield 8
Date:
Nov 2023
Service:
Hackathon
Harry Potter Character Matcher & Prank Call
A 24-hour hackathon project matching a user's face to the Harry Potter character they most resemble, then prank-calling them in character via Twilio. I helped build the training dataset — scraping candidate images per character with Selenium, then filtering, background-removing and normalising them to reduce colour bias — before training a classifier on Teachable Machine. The site (HTML/JavaScript on a Node.js server) takes an uploaded photo or webcam capture, runs it against the model, and triggers a Twilio call scripted around the matched character. Even this early on, our hackathon strategy was the same one we'd repeat in later years: reach for as many different technologies as the 24 hours would allow, treating it as a learning opportunity first. Won Best Hack 3rd place at HackSheffield 7.
Client:
HackSheffield 7
Date:
Nov 2022
Service:
Hackathon
Robotics — Simulated TurtleBot3 Navigation
A second-year Robotics module (COM2009), programming ROS1 nodes to control a simulated TurtleBot3 Waffle in Gazebo — sensing, movement control and reacting to a numbered-zone arena, built around the course's own custom message, service and action definitions. Lecture material covered PID control, sensor fusion, localisation, path planning and SLAM. The screenshot here is genuinely live: I rebuilt the module's original ROS package and relaunched the exact same simulation arena the course used, rather than reconstructing it from memory (65%).
Client:
University of Sheffield
Date:
Jan 2022 - May 2022
Service:
University
My First Portfolio Site
The predecessor to this very site — my first proper personal portfolio, primarily built over an intensive week in August 2021 (with the Sheffield Ethical Student Hackers project page following that November). Built on Jekyll, starting from the jekyll-theme-cayman base and reworked with my own SCSS and JavaScript: a hand-rolled hamburger menu with no library, several passes of mobile-first responsive fixes, a custom image slideshow, and a dedicated CV page. It also handled its own GitHub Pages CNAME/HTTPS setup to serve the AnyQueue project on its own subdomain. Kept here as a record of where the current site (and my front-end skills) started.
Client:
Personal Project
Date:
Jun 2021 - Nov 2021
Service:
Personal Project
The AnyQueue Music System
My first personal web development project, built in Ruby/Sinatra using the Spotify API. AnyQueue lets Spotify Premium users create shareable rooms where guests can add songs to the host's queue via a custom search menu, with all user-added songs visible on the main page. Originally prototyped as a smaller C# version that only let people add to my own queue, I rebuilt it in Ruby for full-stack database management once I saw room to expand it. All the styling is hand-written CSS with no Bootstrap or Tailwind — I hadn't touched a CSS framework yet at this point. Hosted on AWS, currently in Spotify Development mode for beta testers.
Client:
Personal Project
Date:
June 2021
Service:
Personal Project
The eMentor WebApp
My first-year COM1001 (Introduction to Software Engineering) group project — a 7-person team reduced to 4 active contributors due to the pandemic, still delivering a solid product for our client through weekly meetings. Built in Ruby Sinatra, my first real full-stack build. Security wasn't a formal requirement, but I applied hashing and salting, non-consecutive IDs, random cookie tokens, and XSS prevention anyway. I made roughly two-thirds of all commits across the project, which scored 85% in the first semester and 100% in the second, for a 92% module total.
Client:
University of Sheffield
Date:
Dec 2020 - May 2021
Service:
University
Sheffield Ethical Student Hackers Society Website
As Inclusions Officer for the Sheffield Ethical Student Hackers Society, my role extends beyond inclusions into general committee work — helping design and present Ethical Hacking sessions, and making website changes as needed. Committee contributions include fixing formatting on the committee page and building a fully functional hamburger menu using git and multiple branches, a technique I later reused on this site.
Client:
Sheffield Ethical Student Hackers Society
Date:
Ongoing
Service:
Committee Role
The eParent Revision Monitor
My A-Level Computer Science NEA, where using external libraries was actively discouraged to maximise marks for complex, self-built behaviour — so I went beyond the brief and wrote my own RSA encryption for client-server communication. Users create an account and get quizzed with tailored revision questions during pre-set study sessions, enforced by background program detection that blocks distractions (games, browsers) until the session ends. No public release, but I plan to rebuild a cleaner, more polished version eventually.
