Craftivion - Portfolio

IoT Development - C++ with UWB

This project specifically entailed using a multitude of ESP32 microcontrollers to provide positional data to one-another to establish a triangulation system allowing the capability of positional estimation within a virtual environment. This would use ultra-wideband in the form of DW1000 modules, coded in Arduino using C++ to transmit distance estimation between two anchors and a tag, of which this data would then be transmitted via WebSocket to a server in order to calculate the real world position of the tag relative to the anchors and output the result in a visual format on a webpage with real-time updates. Due to limitations at the time, this project was only capable of running in a 2-dimensional environment due to time constraints and more resources being required, further intentions would have led to a 3-dimensional environment within virtual reality with the UWB tag being paired with an IMU and a third anchor, allowing for precise tracking within a virtual environment in both position and orientation.

Android Music Player

The music player on android was another project designed by Donny Deevey which would explore the capabilities of media streaming within the android environment, developed within Java via Android Studio, this would allow the user to grant permissions for the application to access photos and media on the device, upon granting would be provided to a basic user interface which would allow for the playing, pausing, skipping or rewind of any media found on the device. This application would also make use of the android activity system, with the first activity providing input for the current media, with a second activity page to select which media could be played, essentially forming a library for the application to access.

Tojahaca

Tojahaca is a custom multi-server Minecraft network designed to offer diverse gameplay experiences within a single connected ecosystem. We were tasked with developing a proxy-based infrastructure enabling seamless player movement between servers, and implemented core systems including permissions, commands, custom currency, levelling, and server-side custom items to deliver a modded-like experience without client downloads. We then designed an in-game GUI store allowing players to purchase items, cosmetics, and permissions via custom event-driven systems. With the lobby system complete, we would then be tasked with developing a Roman-themed gamemode called "Gladiator" that added significant depth to gameplay by introducing a fully narrative-driven progression system set in the ancient Roman world. Players advanced through a structured story, triggering events based on progression while engaging with custom mobs, automated spawn systems, and advanced combat mechanics, including battles against mythological gods.

Famallama

Famallama is a Minecraft server built to support players with neurological conditions, with a strong focus on accessibility, inclusivity, and ease of use. We were commissioned to design and develop a suite of custom plugins tailored to the server’s needs. These included advanced chat moderation systems, an automated announcement framework, a dynamic firework display system, and multiple gameplay enhancements aimed at improving usability and player experience. All custom plugins were developed using Java, leveraging the Spigot API and NetMinecraftServer (NMS), alongside established libraries such as NBTAPI, PlaceholderAPI and ProtocolLib. This stack enabled us to deliver a concise, high-performance solution aligned with the client’s expectations. A major component of the project involved delivering seamless cross-platform compatibility between Java and Bedrock Edition players. This was achieved through the integration of GeyserMC, Floodgate, and ViaVersion, alongside our custom-built systems. To ensure consistency across platforms, we also utilised a resource pack conversion system, enabling Bedrock players to correctly render custom items originally designed for Java Edition—resulting in a unified experience for all players.

WebXR - Spatial Campsite

The Spatial Campsite project served as a practical demonstration of the capabilities of Babylon.js, showcasing how advanced web-based rendering techniques can be used to create immersive environments. The scene was constructed using features such as radial cloning to efficiently generate natural elements like trees and rocks, alongside a dynamic day–night cycle to simulate the passage of time. Lighting played a key role in enhancing realism, with individual light sources applied to elements such as the campfire and the sun. These sources dynamically interacted with the environment, adjusting based on the time of day to create a more believable and visually engaging experience. Through this project, we demonstrated that WebXR is capable of delivering a level of immersion and spatial depth comparable to native applications. It also highlights that both approaches—web-based technologies using JavaScript frameworks like Babylon.js, and native development using platforms such as Unity—offer distinct advantages. The optimal choice ultimately depends on the client’s requirements, whether prioritizing accessibility and cross-platform deployment or leveraging device-specific SDKs for more specialized functionality.

Unity - Shared Spatial Anchors

Shared Spatial Anchors is an application developed in Unity using the Meta Quest API, designed to enable a synchronized multiplayer mixed reality experience. A networking system is implemented to allow multiple users to join a shared session. Once connected, a spatial anchor is created and distributed across all participants. Upon receiving the anchor data, each user’s environment is aligned relative to the shared anchor point, ensuring consistent spatial positioning between all players. This alignment allows users to perceive and interact within the same virtual space, creating a cohesive shared experience in a real-world setting. The virtual environment consists of a compact outdoor scene featuring terrain and scattered trees to simulate a natural setting. Within this space, users can interact with custom block-based objects—placing them within the environment or passing them between one another—encouraging collaborative and intuitive interaction.

WebXR - Colocation

This project demonstrates the capabilities of WebXR when integrated with complementary technologies, particularly artificial intelligence techniques in computer vision, to enable a co-located shared experience within a web browser. The system leverages external cameras to capture positional data, allowing users to interact within a unified spatial environment. The WebXR application visualizes pose data collected from three external cameras, which is used to calculate real-world positioning through triangulation. The architecture is divided into two components: a client-side application and a server-side backend. Computationally intensive processes, such as triangulation are handled on the server to reduce the performance burden on client devices. By offloading these operations, the client only needs to receive a continuous stream of processed positional data via WebSockets. This approach ensures consistent performance and broad accessibility, enabling the application to run on a wide range of devices without requiring advanced hardware capabilities, while still delivering an accurate and responsive spatial experience.

XR Photo Gallery

The XR Photo Gallery was a commissioned project designed to immerse users in a curated collection of photographs captured by the photographer. Images were stored on the server side and dynamically streamed to the client within a virtual environment, where users experienced smooth transitions between photos rather than viewing static displays. This approach created a more engaging and fluid visual journey, enhanced by background music to establish a calming and atmospheric experience. Although the option to use a single 360-degree image was available, we followed the client’s preference for a grid-based layout. This decision allowed us to utilise standard images within the virtual space, avoiding the technical constraints and specialised requirements associated with projecting 360-degree imagery in a spatial environment. The result was a flexible and visually cohesive system that effectively balanced performance, accessibility, and user experience.

This website

Our website was developed as a demonstration of what can be achieved through a fully manual implementation, as opposed to relying on drag-and-drop website builders. It integrates a range of technologies to ensure ease of navigation, multiple forms of user interaction, and a backend system capable of handling form submissions and requests efficiently. Built using modern frameworks and libraries, the site delivers a fast, responsive, and reliable user experience. Key technologies include jQuery and Bootstrap for front-end interactivity and responsive design, Node.js for backend functionality, and CAPTCHA integration to enhance security and protect against automated submissions. Together, these components create a robust, scalable, and user-focused platform.

Famallama

Famallama requested the implementation of a range of scripts to enhance the usability and accessibility of their website for users interested in joining their server. As part of this, we integrated a chatbot system designed to handle user enquiries in real time, providing immediate assistance for common questions related to server access and onboarding. Where necessary, the system also enables administrators to review and respond directly to user queries, ensuring that more complex issues can be addressed efficiently and appropriately. For more information regarding famallama, visit their website at famallama.co.uk.

Tojahaca

Our work on the Tojahaca website is currently ongoing, and we are excited to share the progress we have made so far. We look forward to showcasing the finished product soon, so please keep an eye on this page for future updates coming shortly.

RealSense 3D Reconstruction

Our 3D reconstruction solution leverages a single Intel RealSense D405 depth sensor to capture high-resolution segments of a target object. These segments are acquired through a Python-based client application and transmitted to a backend web server for processing. The server, built using Node.js, handles computationally intensive tasks by orchestrating child processes that execute dedicated Python scripts. These scripts perform the core 3D reconstruction, combining the captured segments into a unified point cloud model. Following reconstruction, advanced denoising techniques—such as Principal Component Analysis (PCA)—are applied to eliminate noise and irrelevant data, resulting in a cleaner and more accurate representation of the object. The refined point cloud is then stitched and optimized for visualization. Once processing is complete, the server delivers the reconstructed model to the client via a web interface powered by Babylon.js. This enables users to interact with the fully rendered 3D model within a virtual environment, facilitating detailed analysis and allowing for insights that may not be apparent through conventional 2D imagery.

BlazePose tracking implementation

Our implementation of BlazePose leverages its ability to track multiple individuals simultaneously, enabling the capture of positional data mapped directly onto web-based canvases. This facilitates a top-down tracking approach, where the canvas mirrors the dimensions of a real-world space. A key advantage of this method is its cost efficiency—requiring only multiple external video sources, such as standard webcams, rather than expensive dedicated tracking hardware. While higher camera resolutions can improve accuracy, the system remains a viable and accessible alternative to traditional tracking solutions used in virtual reality environments. One of the primary challenges encountered was maintaining persistent and reliable tracking to ensure a continuous stream of positional data. To address this, multiple points of interest are assigned to each individual rather than relying on a single marker. This redundancy allows the system to sustain tracking even when certain points become temporarily occluded, seamlessly falling back to alternative reference points and preserving overall positional integrity.

Clustering Comparison

Our clustering algorithm enables a direct comparison between different clustering models applied to the same dataset, while also allowing the use of different dimensionality reduction techniques. This makes it possible to clearly evaluate the impact of both the chosen model and the transformer on the resulting cluster structure. The system is built with a Tkinter-based user interface for ease of interaction and usability. It supports clustering models such as K-Means, DBSCAN, and OPTICS, alongside transformation methods including PCA, t-SNE, and UMAP. These transformers are used to reduce noise and project the data into a more interpretable form, helping to produce clearer and more visually meaningful cluster arrangements.

Our Work

Click to view

Click to view

Click to view

Click to view

Want to know more?

Our Work

Click to view

Click to view

Click to view

Click to view

Want to know more?

Software Engineering

IoT Development - C++ with UWB

Android Music Player

Tojahaca

Famallama

Spatial Computing

WebXR - Spatial Campsite

Unity - Shared Spatial Anchors

WebXR - Colocation

XR Photo Gallery

Website Development

This website

Famallama

Tojahaca

Artificial Intelligence

RealSense 3D Reconstruction

BlazePose tracking implementation

Clustering Comparison