Led independently from concept to public exhibition, spanning research, interaction design, and technical development. Responsibilities included conducting and coding interviews in Atlas.ti; designing and building the real-time interaction system in TouchDesigner, covering face detection, data visualisation, and scene control; developing custom Python scripts to parse and display live POPS data; sourcing and processing GIS data in QGIS for the companion website; conducting a formal risk assessment and implementing privacy safeguards for the live camera feed; producing visual assets in Figma and After Effects; resolving technical performance issues during installation; and gathering visitor feedback on-site. Methods included media and discourse analysis, ethnographic observation, semi-structured interviews, qualitative coding, secondary and GIS data analysis, iterative prototyping, risk assessment, and in-the-wild user testing. The project ran over three months using TouchDesigner, Python, Figma, After Effects, DaVinci Resolve, QGIS, Atlas, and Atlas.ti.

Secondary research, including academic sources, journalism, and a since-restricted public POPS dataset, was combined with interviews and site observation to identify three recurring dynamics: the gap between perceived and actual publicness, behavioural self-regulation in the absence of visible rules, and surveillance as an atmospheric rather than purely technological condition. Interviews were coded in Atlas.ti concurrently with data collection, allowing early findings to inform subsequent interview questions. These findings converged on a core interaction: a mandatory face scan prior to entry, directly informed by CCTV privacy policies at sites such as King's Cross Estate, which disclose the capture of attributes including clothing and hair colour without users' awareness. The interaction was built end-to-end in TouchDesigner, incorporating face detection, a command-line-style interface displaying live POPS data such as owner, borough, and coordinates via custom Python scripts, and a restriction sequence using custom Google Earth and Maps footage. DaVinci Resolve and After Effects were used interchangeably for map tracking depending on task requirements, and the TouchDesigner network was restructured into modular scenes to support ongoing iteration. The project was initially scoped across six POPS sites, then reduced to three, Elephant Park, Canary Wharf, and King's Cross, based on interview and observational data, prioritising depth and clarity of experience over breadth of coverage. Sound was developed as a core component of the experience rather than a placeholder element, and the visual language was informed by Ryoji Ikeda's data-driven aesthetic and Apple's Face ID interface, establishing familiarity and authority while improving the legibility of dense information. Given the use of a live camera feed, a formal risk assessment addressed consent and data misuse. Safeguards included activation only on explicit user action, local processing, no storage or transmission of footage, and immediate discarding of facial landmark data, supported by a visitor-facing information sheet to ensure transparency. A late-stage hardware performance issue was also resolved by downscaling resolution and implementing custom pixel-format and interpolation handling in Python, stabilising playback for public exhibition.

On-site observation and visitor conversations indicated strong engagement with the face-scan and denial interaction, though the underlying POPS data was frequently under-noticed, with visitors requiring light prompting to locate the accompanying QR code. This provided a clear, actionable insight into where the current design prioritises experiential impact over information delivery, informing potential future iterations. The project demonstrates the ability to move from qualitative research through to full technical execution, including real-time interaction design, data pipelines, and live installation troubleshooting, while maintaining rigour through interview-based evidence and iterative user validation. The refinement from six sites to three also reflects a broader design principle applied throughout: treating scope reduction as a deliberate legibility decision rather than a limitation.