Project · Hackathon · August/September 2018
GachaBot
A physical recommendation machine that decides which few things to ask a tourist, and which to read from the environment — location, weather, traffic, time — to generate a recommendation. Designed, prototyped, and tested with real tourists in 48 hours at Kyoto Startup Summer School (KS³).
Context
What do you do when you have unexpected free time in Kyoto?
You could search Google or check TripAdvisor, but you’ll get the same ranked list of popular attractions everyone else gets. What if you wanted something off the beaten path? Something only locals know about, matched to right now: your group, how much time you have, what the weather’s doing?
That tension between the desire for a unique local experience and the friction of finding one on demand was the problem we set out to investigate during the end-of-program hackathon at Kyoto Startup Summer School (KS³) in September 2018. 48 hours. Four people. No prior knowledge of what the product would be.
Kyoto Institute of Technology · Kyoto Startup Summer School (KS³) · August/September 2018 · the 48-hour hackathon was the program’s final event
GachaBot is a physical recommendation machine modeled after the gacha capsule vending machines already familiar across Japan — a child can operate one, and tourists understand it immediately. The interesting design decision was not the gacha form factor. It was where to draw the line: which few things to ask a person, and which to infer from location, weather, traffic, and time. We had a 30-second target for the full interaction — from first question to capsule in hand — so every question had to earn its place.
Research approach
48 hours, all in the field
We didn’t design in a room. On Saturday afternoon we went to Kyoto Station, Nishiki Market, and tourist information centers with an interview guide and prototypes, and talked to real tourists about their travel planning habits, their frustrations, and their reactions to our concept.
Post-research debrief session
Synthesis and affinity mapping
What tourists told us
Unexpected was exactly what they wanted from GachaBot
The research validated the core insight quickly: tourists found it stressful to identify unique experiences on the fly, and they were actively looking for something that could surface the unexpected. The gacha form factor (familiar, playful, instantly understood) landed immediately.
- Tourists wanted unexpected recommendations, not ranked lists.
- Navigation info on the slip was critical: knowing about a place isn’t useful if you can’t get there.
- The question flow had to stay short. If it took more than 30 seconds, interest dropped.
The product
A gacha machine that knows what’s happening right now
GachaBot is a physical recommendation machine modeled after the gacha capsule vending machines already familiar across Japan. You answer three quick questions, put in 100 yen, and get a capsule with a recommendation slip: a photo, a brief description, and directions to a local experience you wouldn’t have found on your own.
Team building the physical machine
Recommendation slip · v1
Handwritten paper slip from field prototype
Recommendation slip · v2
High-fidelity concept: photo, why it’s worth seeing, directions, and QR code
How it works
Three questions + environmental data → one recommendation
The design logic: the recommendation was not random. The system combined the three answers with ambient signals like location, weather, traffic, and time, so a rainy afternoon would not return an outdoor park, and a late hour would not return a place about to close. We demonstrated this logic with an interactive Axure prototype and paper props.
Three questions — time available, group size, children in the group — combined with environmental data the machine reads automatically. Every question we kept was there because field research showed it changed the output in a way tourists cared about.
The three questions, written on sticky notes during the working session.
Design rationale
Why a machine and not an app
A typical gacha machine wall in Japan — familiar to locals, immediately understood by visitors
We could have built an app. We chose a physical machine on purpose. A gacha capsule machine is understood instantly in Japan, a child can operate it, and it lives exactly where the question comes up: inside the station, the information center, the spot where someone is standing around wondering what to do next. An app has none of that. You have to know it exists, download it, and remember to open it at the right moment.
The physical form also matched what tourists told us they wanted, which was to stumble onto something rather than scroll a list. An app was envisioned as a later expansion — something a tourist might download after the machine had already caught their attention at the point of decision.
What this demonstrates
Research discipline under time pressure
GachaBot compresses a full product cycle into 48 hours, but the research didn’t take shortcuts. The field findings directly changed the design the same day they were collected.
| What this shows | How it showed up at KS³ |
|---|---|
| Deciding what to ask vs. what to infer | The hardest design question wasn’t the recommendation algorithm — it was the question flow. With a 30-second target for the full interaction, we had to decide what a person genuinely needed to answer, and what the machine could read from context (location, weather, traffic, time) without asking. Every question we kept had to justify being there. |
| Research that changed the product same-day | We were in the field Saturday afternoon and had findings reflected in the prototype by Saturday evening. The “Are you traveling with children?” question wasn’t in the original design — tourists told us group composition mattered, and it went into the next prototype on the spot. |
| Design decisions grounded in research, not assumptions | The 3-question limit and the navigation info on the recommendation slip trace directly to what tourists told us in the field. The gacha form factor was a hypothesis we developed before going out — the research confirmed it landed. |
| Field research without infrastructure | No screener, no lab, no preparation time beyond that morning. We built an interview guide, went to Kyoto Station and Nishiki Market, and talked to real people with prototypes. |
Related work
Self-guided cooking skill app: 7 interviews, 2 design rounds, review of 8 products
Enterprise intranet IA, user research, and interaction design for 45,000+ employees
A home-decor brand run almost solo: a study in where AI helps and where humans decide