Work / KT AI Station
KT AI Station
Three AI ethics programs for young people. Visitors watch a deepfake made from their own face, pick out AI-made news and find their digital citizen type.
- Client
- KT
- Where
- Touring schools in remote regions · Seoul Robot & AI Museum
- When
- 2025 — 2026
- Our part
- Software development and every interaction
WSIS Prize 2026 Champion · Information and knowledge access (Digital Daily, 12 Jul 2026)
KT AI Station is KT’s mobile AI education platform. It visits schools in remote regions so young people can experience AI technology and AI ethics first-hand, and it runs permanently at the Seoul Robot & AI Museum. In July 2026 it was named a Champion of the ITU’s WSIS Prize in the information and knowledge access category. PUTURE built the software for its three ethics programs, the GPU face-swap server and the session booking system.
- 1,797 young people in its first year, in 21 regions and 19 schools (Digital Daily, 12 Jul 2026)
- 1 / 1,595 projects; the only Korean winner of WSIS Prize 2026 (Asia Economy, 12 Jul 2026)
- 4.77 / 5 student satisfaction, 4.94 from teachers (Digital Daily, 12 Jul 2026)
Deepfake
Visitors photograph their face at the kiosk and pick a scene card. About 30 seconds later a film plays with that face swapped in, followed by how deepfake harm spreads and how to prevent it.
Key engineering
- Touch-kiosk web app with personal-data consent and camera-capture flows
- Face detection, swap and restoration pipeline on a cloud GPU instance (12 scenario films, rendered in about 30 seconds)
- Asynchronous render-job API (job submission, status polling, streamed result), with the captured photo and output video deleted from the server when the visitor finishes
- Serverless control plane that starts and stops the GPU instance on the operating schedule or by messenger-bot command, and updates the kiosk config only after a test face swap passes
Fake news
Two news reports play, and visitors pick the one made with AI or answer that both are fake. An answer film follows, then a comparison with what others chose for the same pair.
Key engineering
- Random two-of-six pairing of news reports (never the previous pair), each pair mapped to its own answer film
- Responses logged to a local database, with the choice split aggregated per pair in real time
- Keyboard shortcuts for operators: reset to start, video skip, full-screen toggle
Digital Citizen Type Test (DCTI)
Twenty-seven picture items assess how visitors think, learn and act online and sort them into one of eight digital citizen types. Results come as a type card, a matching job and per-domain scores.
Key engineering
- Picture-based forced-choice assessment interface: 27 items across three domains
- Scoring logic that assigns one of eight types (2×2×2) from the answer split in each domain, with a result card and ratio charts
- Admin console on a local database: stored responses, per-trait counts and CSV export
The DCTI model was developed by Ewha Womans University with KT.
Key engineering
System
- Touch screen
- Kiosk camera
- Booking QR
- Three program apps
- GPU face-swap server
- Local response database
- Deepfake film
- Type card
- Live statistics
Web kiosk UI, Motion graphics, Face-swap models, Cloud GPU instances, Serverless workflows, Local database, REST API
Visitors book sessions through PUTURE’s on-site booking system