AI Lab — H2O Inc.

INTENT
OVER
ALGORITHM

In an era where AI evolves at an accelerating pace,
don’t get swallowed by AI.
We help you execute AI initiatives that honor human intent.

View projects
3.9×
GenAI adoption
(DAU comparison)
▲83%
Lead time reduced
24 months → 4 months
14×
Dev efficiency (LoC)
1-year results
2日
90% of mortgage cases
in 2 business days (was 14)
INTENT
OVER
ALGORITHM
Vision

Our philosophy

Don’t get swallowed by AI.
We help you execute AI initiatives
that honor human intent.
Intent Over Algorithm

As technology advances at an extraordinary pace worldwide, many organizations are moving to adopt AI. Yet too many initiatives stall at the PoC stage and never reach production.

AI Lab helps you identify where AI creates real value, and supports the design and implementation required to integrate human thinking, intent, and strategy with AI in the right way.

With “technological resilience” as our foundation, we help frontier technology take root in real operations—without waste, and with clear ownership.

Services

Three services

01 — Advanced R&D
Advanced R&D

We create frontier value with top-tier researchers in machine learning and natural language processing (Ph.D. level). We deliver leading capabilities in document intelligence, NLP, AI agents, and cognitive science.

NLP / LLM AI Agent Cognitive Science RAG
AI
02 — Product
Product delivery

A collaborative team of former CTOs and AI-focused PMs enables smooth, fast delivery. We implement even from ambiguous requirements by understanding business structure, and ship quickly with a prototype-first approach.

Prototype Rapid delivery CTO support
DX
03 — Consulting
AI consulting

From understanding operations to implementation, we work with you to execute what truly matters. With an AI-UX-first design methodology, we help AI become embedded in the organization—so DX doesn’t become performative.

AI-UX DX strategy Adoption
CX
Members

Members

A small, specialist team that understands both business and technology.

The team is centered on members with frontline experience in AI research, founding companies, and large-scale product development. Specialists skilled in moving projects forward and implementing LLMs and AI agents take part alongside them.

From framing the problem while requirements are still vague, to rapid prototyping, to building systems that hold up in production, specialists in each field stay with the work throughout.

It is a team that combines the agility a large organization does not have with frontline expertise.

Projects

Project track record

Here are three representative examples.

Challenge

  • Sales thinking became individualized as customer problems grew more complex.
  • Knowledge was fragmented, preventing consistent proposal quality.

Approach

  • Designed for low-friction, everyday usage by sales teams.
  • Embedded frameworks (account planning, differentiation, etc.) so the Copilot supports strategy—rather than simply “answering.”
  • Integrated across sales moments with RAG in the background.

Impact

DAU 19.5%3.9× higher usage vs. the internal GPT tool
Faster proposalsMultiple cases of dramatically faster proposals than the baseline
Knowledge compoundingRapid knowledge accumulation observed from beta

Challenge

  • Development requests were centralized in the DX unit, creating operational overload.
  • Moving from PoC to production often took more than two years.

Approach

  • Designed a “DX as a Service” foundation with common APIs, LLM, and RAG.
  • Built an agent catalog (20 types): chat, RAG, workflow integrations, and more.
  • Enabled business units to deploy independently via no-code prompt editing and RAG management.

Impact

24M → 4MLead time reduced by 83%
¥300M+/yrEstimated annual cost savings at beta launch
Parallel AI rolloutEnabled AI adoption at a scale difficult to replicate

Challenge

  • Data fragmentation inflated touch time across mortgage applications, financial statement analysis, and KYC.
  • Regulatory audit logs were manual and costly; many PoCs couldn’t reach production.

Approach

  • Graph-modeled mortgages, financials, and KYC (240M nodes / 12M evidence links).
  • Applied LLM + rules for field extraction and anomaly detection, auto-attaching explainability chains.
  • Built dynamic HITL orchestration that adjusts review depth by risk tier.

Impact

14d → 2d90% of applications processed within 2 business days
Global referenceAdopted as a reference case across worldwide branches
Design your AI project,
together.

Don’t stop at PoC. Build AI that sticks in real operations.
Reach out anytime—we’d love to talk.