Who We Are and What We Ship

EFFOMA builds AI agent systems and extended reality solutions. Our agents process foreign resident registrations at Japanese municipal offices, guide tourists at rural railway stations, and help patients navigate hospital check-in procedures. They run on digital signage and touchscreen kiosks with integrated hardware — IC card readers, QR handoff to mobile devices, and real-time staff alerting systems.

We also build VR simulations — automotive showrooms, architectural walkthroughs, and enterprise data visualization.

We operate from Ho Chi Minh City (engineering), Sydney (international HQ), and Tokyo's Minato ward (Japan operations and partnerships).

!ARISA AI agent deployed at exhibition booth

What Is Live in Production

These systems are deployed and serving real people today.

RIN handles foreign resident registration at Myoko City Hall in Niigata Prefecture. When a foreign worker arrives to register, RIN reads their residence card via IC reader, auto-populates government forms, and walks them through tax, pension, health insurance, and municipal procedures — in their language, with explanations adapted to their home country's administrative systems. Registration time dropped from 30-60 minutes to under 10.

ARISA and ALICE are AI tourism guides deployed at Seibu Railway stations in the Chichibu region. They run as 3D avatars on station signage, providing restaurant recommendations, hiking trail conditions, temple hours, and train schedules. When a visitor scans a QR code, the conversation transfers to their phone and stays with them for their entire trip.

SOFIA assists patients in healthcare facilities with scheduling, check-in, navigation, and insurance explanation. She never provides medical advice; medical questions go to a nurse immediately, with a conversation summary attached.

All four agents are proactive — they initiate conversations based on context and behavioral patterns, not just respond to queries. They maintain conversation history across sessions and trigger real-time staff alerts when they detect situations requiring human intervention.

!Seibu Railway AI agents — ARISA and ALICE at station deployment

Why We Build on Claude

Our platform, Agentfront.ai, uses Claude by Anthropic as its core reasoning engine. This is a deliberate technical choice, informed by the current state of frontier models.

Here is where the major models stand as of March 2026, based on publicly verified benchmarks:

Frontier Model Landscape (March 2026)

Sources: Chatbot Arena (lmsys.org), SWE-bench verified leaderboard, individual model documentation. Benchmarks are point-in-time snapshots; model capabilities evolve rapidly.

Each model has genuine strengths. GPT-5.4 excels at coding tasks and IDE integration. Gemini 3.1 Pro offers unmatched context length at competitive pricing. Kimi K2.5 is the strongest open-source agentic model, with impressive autonomous task execution.

We chose Claude for three specific reasons relevant to our deployment context:

1. Japanese formality handling. Our agents operate in government offices where keigo (formal Japanese) is non-negotiable. Claude maintains correct formality registers across multi-turn conversations better than alternatives we tested.

1. Concept mediation, not just translation. When a Vietnamese resident asks about Japanese pension enrollment, Claude explains the concept relative to Vietnam's BHXH system — not just translates the words. This cross-cultural knowledge depth is the difference between a useful agent and a confusing one.

1. Safety boundary respect. In government and healthcare, the model's instinct toward caution is an asset. Claude consistently respects scope boundaries and escalates to humans when conversations exceed the agent's authorized domain.

Claude is the right fit for high-stakes, multilingual, safety-critical deployments. That is what we build.

Our VR and XR Portfolio

!VR automotive showroom — immersive car buying experience

- Automotive VR showrooms with real-time vehicle customization - Architectural walkthroughs of buildings before construction begins - Simulation environments adaptable for training and education use cases - Enterprise 3D data visualization for collaborative analysis

!Architectural visualization in virtual reality

We are developing AI-enhanced XR experiences — virtual training instructors that adapt to the learner, natural language interfaces for navigating 3D environments, and context-aware guides for architectural and urban planning simulations.

What You Will Find on This Blog

We write about what we encounter building and deploying these systems. Not thought leadership. Not predictions. Just what happened, what we learned, and what we are building next.

Start here:

- What Breaks When You Ship an AI Agent to a Japanese City Hall: Engineering lessons from production deployments. What goes wrong, what we got wrong, and the principles we built from it. - From 60 Minutes to 10: AI Agents at Japanese Government Offices and Railway Stations: A detailed case study of our deployments and what multilingual AI actually means when a Filipino resident needs to understand Japanese pension law. - The 2026 Reasoning Race: Kimi K2.5, Claude, and GPT-5 in Practice: A neutral comparison of frontier models from a practitioner perspective.

Subscribe to our RSS feed or check back regularly.

---

Questions, partnerships, or just want to talk: [email protected]