Almost Kin
NOA reads you and answers.
A fictional AI from 2970 scans gallery visitors via webcam and returns a personalised line generated from their current state. Shown at the Kinship exhibition, Safe House Gallery 2, Peckham.
Client
Self-initiated
Year
2026
Industry
Interactive Media
Turnaround time
~6 weeks

The Question
Most emotion-recognition systems point outward, they classify people to sell them things. I wanted to ask what happens when the direction reverses: what if a machine used the same technology to give something back? Not a recommendation, not a metric, a Line. Something intimate, non-utilitarian, and slightly uncanny in the way only a machine's attention can be.
Almost Kin was built for ENTANGLED, a group exhibition at Safe House 2 in Peckham. The brief was open. The constraint I set myself was that the piece had to make a visitor feel noticed rather than scanned.
The Approach
The work is a three-part system. A webcam feeds MediaPipe face landmarks into a Random Forest classifier trained on five emotional states — nostalgia, comfort, uncertainty, anticipation, and emotional distance. The classifier's output is passed via OSC into a Unity front-end styled as NOA, a fictional AI from the year 2970. NOA reads the visitor for around eight seconds, then displays a poem selected from a hand-written library, one for each classified state.
The tone of NOA matters as much as the tech. She speaks like a distant relative who's been reading you your whole life. The visual language is a squircle-grid front-end in dark terminal aesthetic, deliberately restrained so nothing competes with the poem itself.
The Classifier — Random Forest trained on the Limerence dataset, chosen over a neural net for interpretability. Five emotional states, ~87% accuracy in gallery lighting conditions.
The Bridge — Python and Unity communicate over OSC. Python does the classification; Unity handles the interface, timing, and poem selection.
The Lines — Written by hand, one per state. NOA never repeats herself in a single session.
The Outcome
Almost Kin ran for two days at Safe House 2 as part of the ENTANGLED exhibition (26–27 June 2026). Visitors returned multiple times. The most common response was a pause, a re-read, and a quiet exit, which was the response I'd hoped for.
The OSC-plus-Unity pipeline built for Almost Kin has since become the foundation for two subsequent projects, including the kinetic panel extension currently in development.
"It felt like being read by someone who's been waiting a long time to see you."
— Visitor, ENTANGLED 26.06.26
Project Highlights
Real-time facial emotion classification pipeline (MediaPipe + Random Forest)
Custom Unity front-end with squircle-grid interface
Original poem library for each classified emotional state
Python ↔ Unity OSC bridge (reusable across projects)
Exhibited at ENTANGLED, Safe House 2, Peckham






