Background
The Singapore Prison Service (SPS) operated an existing VR rehabilitation programme of eleven fixed scenarios built in Unreal Engine, designed to help incarcerated drug users resist the social pressures that trigger relapse. The programme was effective on first exposure, but its finite, scripted nature created diminishing returns: repeat users came to anticipate each scenario, undermining the desensitisation effect the programme was designed to deliver.
The deployment
LIVR LTD, a UK-based AI and immersive technology startup, was engaged through HTX Hatch's Dimension X Cohort 3 open innovation programme to address this limitation. Working in partnership with the HTX Human Factors and Simulation Centre of Expertise (HFS CoE) and under the guidance of lead engineer Derek Chong, LIVR developed LIVR AVATARS — a fully immersive VR platform in which photorealistic AI-driven avatars respond dynamically to user speech, behaviour, and physiological signals.
The system generates a unique session every run. Therapists configure scenario frameworks and guardrails; the AI navigates from within those parameters, continuously finding new pressure vectors based on each user's responses. A user who has completed the programme multiple times encounters a different scenario each time, because the system responds to how they present in that moment — their mood, resistance level, and spoken replies.
Technology
The platform runs entirely on-device, eliminating network dependency and ensuring complete data isolation — a non-negotiable requirement in a custodial government setting. AI inference latency is approximately one second. Biometric monitoring captures pupil dilation, heart rate, vocal sentiment, and attentional focus in real time, giving clinical staff a live physiological picture alongside behavioural observation and enabling personalised session management over time. The platform also runs on desktop and tablet, preserving operational flexibility across different institutional settings.
Cultural and linguistic localisation
Cultural accuracy was a significant engineering challenge. The avatar system incorporates Singlish speech patterns, regional accents, and local vocabulary. The initial training environment, modelled on UK drug dens, was redesigned after SPS flagged that elements such as graffiti and broken windows do not appear in Singapore's urban landscape and were actively reducing immersive realism for users.
Outcomes and recognition
The proof-of-concept was completed across a four-month development cycle and demonstrated at Singapore Week of Innovation and Technology (SWITCH) to government stakeholders, venture capital firms, and system integrators. Following a week-long evaluation by SPS across more than 200 government submissions, the project received the HTX Innovation Award. LIVR subsequently opened a Singapore office to support continued development across the Asia-Pacific region.
