Work with us

AI that holds up
under examination.

Defensible by construction — not by retrofit.

We design AI systems for banks, insurers, and other enterprises with cognitive stakes — systems that can show an auditor exactly where every answer came from, hold a boardroom's disagreement on the record, and read the customer they are talking to.

WHY US

Why the Institute

Hajime Institute is small by design and unusual by intent. Its director has founded four operating companies in AI — including Cinnamon AI and Nexus FrontierTech, both of which deployed document AI inside regulated enterprises at scale — while continuing to publish peer‑reviewed research at the venues that judge the field.

That dual track is not biography. It is the only methodology under which the engineering of AI‑meeting‑cognition can survive: scientific work disciplined by what runs in production, engineering work disciplined by what defends in peer review. When the Institute engages with a partner, both halves of that discipline are in the room.

What we ship is the strategy: the cognitive‑context audit, the functional metric specification, the severity‑tested prototype, the deployment advisory. What you ship is the system — built by your engineers or a partner firm, with the Institute accountable for the discipline throughout. We do not arrive with a generic AI consulting deck; we arrive with the specific scars and metrics the operating side gave us, and the honesty to tell you when a chosen architecture will fail before it does.

MODEL

How we engage

Led by research. Delivered as consulting and advisory. Implemented by your team or by partners.

We do not staff long implementation teams. We carry the research and the methodology; we lead observation and milestone design as consultants; we advise on integration. The implementation itself is carried by your team, by a partner firm, or by an Institute‑associated operating company.

Ⅰ · Strategy & Observation

Cognitive‑context audit.

We come in early, decompose the cognitive context your AI must engage (the document of record, the deliberating team, the user's mind, the surrounding culture), and design the functional metrics by which success will actually be measured — distinct from the benchmark metrics that hide failure modes.

2–8 weeks. Output: a context map, a metric specification, and a milestone plan.

Ⅱ · Prototype & Severity Test

Build, then audit against failure.

For selected engagements, we build a prototype directly inside the Institute (or in tight partnership with one of our operating partners) and severity‑test it against pre‑registered failure modes. The discipline is to fail visibly before deployment, not invisibly during it.

6–16 weeks. Output: working prototype, severity-test report, refusal logic.

Ⅲ · Deployment Advisory

Integration into operating institutions.

Once your engineering team or partner firm carries the implementation, we serve as ongoing advisor on integration — the persona contract, the deployment‑layer architecture, the audit interfaces, the regulator‑ready evidence trail. The Institute remains accountable for the discipline; you remain accountable for the deployment.

Ongoing. Output: integration playbook, audit interfaces, advisory access.

Ⅳ · Research Partnership

Joint paper, joint discipline.

For partners willing to share data and outcomes under appropriate protection, we co‑author peer‑reviewed publications that make your deployment a founding case study in cognition‑aware engineering. This is the highest form of engagement and the one that survives longest.

12–24 months. Output: a co-authored manuscript submitted to a peer-reviewed venue, and a public methodology contribution.

Working across borders

Incorporated in Singapore, the Institute operates from Tokyo and Hanoi and works with partners wherever the problem lives. Engagements run async‑first in English, with standing sessions scheduled to the partner's hours and milestone reviews held in person. Every deliverable is a written artifact — a specification, a report, a playbook — built to circulate inside your organization without us in the room.

CONTEXTS

Where the work lands

Four cognitive contexts where we research, build, and publish. Each maps to a research program; each can become the foundation of an engagement.

Ⅱ  Trustworthy Documents

The Record — audit, contracts, financial statements, regulatory filings.

Sectors · finance · law · regulated industry · supply chain

Trustworthy document reading at a standard defensible to auditors, lawyers, and regulators. Provenance treated as a hard constraint, not a feature: an output that cannot cite its page, clause, and figure is never produced — so one error stays one error. The Institute's directorship has years of operating deployment in this exact domain (Cinnamon AI, Nexus FrontierTech).

Operating record: years of audit‑defensible deployment across finance, insurance, and regulated industry, through the Director’s portfolio companies.

See the program →
Ⅲ  Deliberative Agents

The Team — strategic deliberation, executive decision support.

Sectors · executive committees · strategy · corporate development

Multi‑agent dialectical systems for strategic decisions whose disagreements cannot be optimized away — preserving the irreducible tensions that define the problem, rather than collapsing them into a smooth recommendation no one in the room can defend. The decision file shows the tensions the committee actually weighed.

Field protocols — decision cockpits, divergence labs, KPI‑tree co‑creation, orchestrated agents — operationalized within strategy functions and executive committees of major enterprises.

See the program →
Ⅰ  Cognition‑aware AI

The Mind — customer-facing AI, coaching, mental health.

Sectors · customer support · coaching · mental-health support · education

Cognition‑aware engineering for the surfaces where AI meets a human mind under pressure. The same instruments — affect calibration, distortion detection, trait-aware personalization, escalation and refusal logic — that support a coaching conversation also govern whether a customer‑support agent de‑escalates a difficult call or compounds it. Engagements in clinical or regulatory contexts proceed only with appropriate clinical advisory in place. The program works on AI that supports human reasoning — not on persuasion-for-sales, which is incompatible with its founding principle (Human Intent over Algorithm).

Functional metrics — emotion‑reasoning fidelity, distortion detection, escalation precision — designed from field observation rather than borrowed from benchmarks. Published in EMNLP, ACII, ACL, ICALT, and PACLIC.

See the program →
Ⅳ  Collective Cognition

The Culture — research collaborations on organizational and cultural cognition.

For research-led engagements with HR & organizational science partners

This program is primarily a research initiative on computational distributed cognition — multi‑agent study of how teams, institutions, and cultures reason together. We engage selectively with partners who are interested in research collaboration on these questions, rather than off-the-shelf diagnostic products.

Research anchors: Organizational Dynamics simulation (CogSci 2026) · a community validation discipline for multi-agent simulation.

See the program →
EXHIBITS

Selected engagements

Described at the level
our agreements allow.

Client work at the Institute is protected by mutual agreement — sector, cognitive context, and the shape of the work are what this page can say.

Exhibit Ⅰ

Telecommunications — the executive room.

Ongoing advisory to the strategy function of a major telecommunications group: decision cockpits, divergence labs, and KPI‑tree co‑creation, operationalized inside executive deliberation rather than delivered as a report.

Program Ⅲ · Deliberative Agents · ongoing

Exhibit Ⅱ

Financial services — the documents.

Years of audit‑defensible document AI in production across finance and insurance, carried through the Director's operating companies — the deployment record on which the Trustworthy Documents program stands.

Program Ⅱ · Trustworthy Documents · operating record

Exhibit Ⅲ

Strategic consulting — the methodology.

Advisory to a strategic consulting practice, bringing the Institute's functional‑metric discipline — the metrics that surface failure modes benchmarks hide — into how AI work is scoped and evaluated for their clients.

Advisory · functional metrics · ongoing

Named references are available in conversation, under mutual agreement — the same protection every partner on this page receives.

RECORD

Operating track record

The Institute's directorship carries an unusual operating record for an AI research organization. Cirius Technologies (founded, acquired by Yahoo! Japan). Naked Technology (founded, acquired by Mixi). Cinnamon AI (founded — Japan–Vietnam document AI, years of enterprise deployment in finance, insurance, and regulated industry). Nexus FrontierTech (founded — global document AI, deployed at financial-services scale for US/UK markets). Methods peer-reviewed at ACL, EMNLP, CogSci, and PAKDD.

See the full track record →

For the formal terms of Sponsored Research — and how it differs from Collaboration and Patronage —

How the Institute works with partners →

For the Institute's full philosophical position behind these engagements —

Read the Director's letter →

ENGAGE

Executive briefing

Request an
executive briefing.

Sixty minutes with the Director on the cognitive contexts your AI must survive — where it is likely to fail, and what the discipline looks like applied to your case.

A note outlining your situation is the best way to begin. The Director responds personally.