V

Chapter V

The Final Year Boss Fight

Project Magpie

"Every great journey has a boss level — mine was Magpie."

During the final semester of my MSc in Computing (Advanced Software Development – Data Science), I collaborated with a team of six to build Magpie, a geospatial intelligence platform designed to empower urban planners and non-expert users alike.

The Problem

City planners often struggle with fragmented, outdated, or overly complex tools to analyze public infrastructure like bike stations, parking spots, and public toilets.

We saw a huge gap — no single platform provided a real-time, visual summary of public amenities with accessibility for non-technical users.

Key Challenges

  • Fragmented data sources across multiple city departments
  • Complex tools requiring technical expertise
  • Lack of real-time, up-to-date information

Fragmented Urban Data

The Solution

Magpie: Services at a Glance

We created a comprehensive web platform providing an at-a-glance overview of public services in Dublinthrough automated detection, interactive maps, and real-world data visualization —all without relying on existing centralized datasets.

Unified Platform

What Made Magpie Special

A powerful combination of AI, geospatial technology, and user-centered design

AI-Powered Detection

YOLOv8-based machine learning for automated detection of amenities using satellite imagery. No manual input required!

Interactive Mapping

Mapbox-based visualization with dynamic, interactive location data and smart amenity search with geocoding via Nominatim.

Secure Infrastructure

Public API secured with JWT-based authentication, containerized using Docker + Kubernetes with CI/CD pipelines.

Robust Backend

Backend infrastructure built in Go with PostGIS for geospatial data processing and scalable performance.

User-Centered Design

UX validated by domain experts in accessibility and urban planning, designed for non-technical users.

Real-time Updates

Dynamic data visualization with automated detection eliminating the need for manual data entry and updates.

My Role: Full-Stack Developer

Contributing to both frontend and backend systems with a focus on user experience and technical excellence

Frontend Excellence

  • React Architecture with TailwindCSS and Next.js
  • Map Interactions using Mapbox GL & Deck.gl
  • UI Prototyping with Shadcn/UI & Radix Primitives
  • Accessibility Focus for non-technical users

Backend & DevOps

  • CI/CD Pipelines with GitHub Actions and Flux GitOps
  • ML Integration for YOLOv8 car detection systems
  • Infrastructure using Docker and Kubernetes
  • Team Collaboration in cross-functional development

Machine Learning Highlights

Our biggest technical challenge was detecting on-street and off-street parking from satellite images

The ML Pipeline

1

Model Training

Trained YOLOv8 model on 19,500+ images of Dublin

2

Classification

Classified detected cars as on the road, parked, or empty spots

3

Data Processing

Converted detections into structured, geo-tagged data using novel road mask algorithm

4

Clustering

Used DBSCAN to distinguish public vs private vs lots

YOLOv8 Detection

Results & Impact

MVP Delivered

Fully working MVP with CI/CD pipelines and dynamic map-based UX

User Tested

Evaluated with 11+ real users including planners and accessibility experts

High Impact

Achieved high satisfaction scores and meaningful feature insights for future work

What I Learned

Beyond the technical achievements, this project taught me invaluable lessons about teamwork, real-world application development, and user-centered design

Team Collaboration

Working in a cross-functional team of six taught me the importance of clear communication, shared ownership, and leveraging diverse expertise to tackle complex challenges.

ML in Production

Integrating machine learning models into real-world applications requires careful consideration of data pipelines, model performance, and user experience design.

Accessibility Focus

Designing for accessibility and non-technical audiences requires empathy, iterative testing, and a deep understanding of user needs and capabilities.

Geospatial Development

Building scalable geospatial applications with Go and PostGIS opened my eyes to the complexities and possibilities of location-based data processing and visualization.

What's Next After the Boss Fight?

Having conquered the final year project challenge, let's explore what the future holds and where my journey in technology is heading next.

Read Chapter VI: What's Next