What Hackathons and Film Festivals Taught Us About Building AI Agents

Before EFFOMA built AI agents for Japanese government offices, we spent a lot of time in hackathons and film festivals. That might sound like an unusual foundation for enterprise AI deployment, but the lessons transferred directly.

Here are the most useful things we learned.

Build for the User in Front of You, Not the User You Imagine

Hackathons force you to present to judges who do not care about your tech stack. They care about whether your solution helps someone. Film festivals force you to create experiences where the technology disappears — the audience should think about the story, not the headset.

This principle directly shapes how we build AI agents. When a foreign resident stands at Myoko City Hall trying to register for pension enrollment, they should not be thinking about the AI. They should be thinking about their pension enrollment. The agent's job is to remove friction, not to be impressive.

!AI agent deployment — real people, real context

Context Beats Cleverness Every Time

At a Google hackathon in Melbourne, the Deloitte partners on the judging panel kept asking the same question: "What happens when this encounters a situation you did not plan for?"

At the time, we did not have a good answer. Now we do: the agent stops and escalates to a human. That is a design decision, not a failure mode. Knowing the limits of your system is more important than expanding its capabilities.

This lesson applies directly to our government deployments. When a resident asks RIN something outside its authorized scope, the agent transfers to staff immediately. No hedging. No partial answers. The staff member gets a full conversation summary so nothing needs to be repeated.

International Experience Changes How You Build

Competing across different countries — presenting in Silicon Valley, Singapore, Melbourne, Hamburg — exposed us to fundamentally different expectations about how technology should interact with people.

In Japan, formality and precision are non-negotiable. In Southeast Asia, warmth and approachability matter more. In Australia, directness is valued. These are observable patterns, not stereotypes, and they affect whether people trust and use the systems you build.

Our agents reflect this. RIN is formal and precise for government contexts. ARISA is knowledgeable and enthusiastic for tourism. ALICE is warm for families. SOFIA is calm for healthcare. These are functional differences that determine whether the deployment succeeds or fails.

!ARISA and ALICE — AI tourism agents adapted for different interaction styles

Immersive Technology and AI Are Converging

Our VR work — automotive showrooms, architectural walkthroughs, enterprise simulations — taught us how to build technology that integrates with physical spaces and hardware. That experience directly informed our approach to deploying AI agents on digital signage, touchscreen kiosks, and IC card readers at Japanese government offices.

The next generation of what we build combines both: AI-powered instructors in VR training environments, natural language interfaces for navigating 3D spaces, context-aware guides that understand both the virtual and physical world.

!VR automotive showroom — where immersive technology meets enterprise needs

The Foundation

Every competition we entered, every festival we submitted to, and every presentation we gave to skeptical judges reinforced the same principle: technology works when it is designed for a specific context and a specific person's needs. Not adapted after the fact. Designed from the start.

EFFOMA is built on that principle. Our agents work at Myoko City Hall and Seibu Railway stations because we built them specifically for those environments, those users, and those challenges.

---

Read more about our AI agent deployments: case study | engineering lessons