DIL — Decision Intelligence Lab
SPEECH AI × LLM — 23 YEARS
Decision Intelligence Lab

23 years at the frontier of speech AI and LLMs, from research to product.
That expertise, behind your decisions.

From humanoid speech processing, spoken dialogue, and signal processing to large language models, multi-agent systems, and real-time speech research in the US — I have spent 23 years on both the research frontier and the business front line. As someone who knows the technology from the inside, I support you end to end: from analyzing your problems to deciding what to adopt.

GET IN TOUCH
Masahito Togami — Ph.D. in Engineering, Founder
01 — USE CASES

Where I can help

You don't know where to start with AI

From taking stock of your problems to prioritizing where to invest — I make "what to do first" clear.

You can't judge whether a vendor proposal is sound

A second opinion on estimates, architecture, and feasibility. Avoid over-investment and costly rework.

You want voice or conversational AI in your product

Design decisions for speech recognition, dialogue systems, and voice UX — what's possible now, and what's still hard.

You want generative AI and multi-agent systems to transform operations

What to delegate to LLM agents, and where to keep human judgment. From concept design through PoC validation to production.

You're struggling with AI hiring and team building

Who to hire, and in what order. Advice on designing and growing R&D organizations, from first-hand experience.

You want an outside perspective on your R&D direction

A sparring partner for your technology roadmap — opportunities and risks, seen from the research frontier.

02 — SERVICES

Three ways to work together

Speech AI & dialogue systems LLMs & generative AI Multi-agent systems Physical AI & robotics Building AI R&D organizations
RECOMMENDED
SPOT

Spot Consultation

90 minutes / online or in person

A second opinion on technical decisions, a sounding board for AI strategy, or an evaluation of vendor proposals — a single, self-contained session.

¥120,000The best starting point for new clients
DIAGNOSIS

Initial Diagnosis

3 sessions + written report

I map your challenges and assess AI feasibility and priorities from a technical standpoint — so you know exactly where to start.

Contact meQuoted based on scope
ADVISORY

Monthly Advisory

2 sessions / month + async support

Ongoing technical and strategic support: steering development, hiring and org building, and hands-on R&D guidance.

Contact meMinimum 3-month engagement

* All engagements are provided under a formal advisory agreement.

03 — APPROACH

Start from the problem,
not the solution

Initiatives that start with "what could we do with this AI?" usually stall before implementation. My approach is problem-driven: first I translate your business challenges into the language of technology, then determine which technologies you should use now — and which you shouldn't.

Connecting problems to the right solutions requires deep knowledge of both. For 23 years I have stood on the research frontier and the business front line at once. Bridging problems and technology — that is where I find my value.

04 — CAREER

Career

  1. 2003 — 2018
    Hitachi, Ltd. — Research & Development Group

    Research in source separation and speech enhancement. Developed the speech processing for the humanoid robot EMIEW. Led the speech and signal processing team. From 2016, drove joint research with Stanford University and others at Hitachi America.

  2. 2018 — 2021
    LINE

    Principal Researcher, then Senior Manager leading an AI R&D organization.

  3. 2021 — 2024
    Amazon Web Services (US)

    Principal Applied Scientist. Led real-time speech analytics R&D on the Chime SDK team, commercializing Voice Tone Analysis and more.

  4. 2024 — 2026
    Amazon Japan

    Principal Applied Scientist / Senior Manager. Led generative AI projects in Japan and drove multi-agent R&D.

  5. 2026 —
    Decision Intelligence Lab — Founder

    Independent. Supporting enterprise AI adoption as a technical advisor.

  6. 2026 —
    Waseda University — Visiting Researcher

    Research on intelligent meeting agents and collective decision-making at the Perceptual Information Systems Lab.

Author: 『Pythonで学ぶ音源分離』 (Source Separation with Python) — Impress, Machine Learning in Practice series GOOGLE SCHOLAR ↗ DBLP ↗
75+
Peer-reviewed papers
20+
Patent applications
IEEE
Senior Member
Ph.D.
Engineering
05 — INSIGHTS

Writing & talks

BLOG · BUSINESS

AI decision support on data you can't send outside — letting a smart AI pave the way for a weaker one

How to get AI decision support from confidential data without sending it to a cloud service. A powerful cloud model designs the system architecture; a weak local LLM then executes it inside your secure environment. The post also walks through a concrete design example built around a household-finance simulator demo. (Japanese)

note · 2026.08.27READ ↗
 TECHNICAL DEEP DIVE · ZENN Let the smart one pave the way: non-blocking design for systems containing local LLMs Zenn ↗
BLOG

Designing the "field" that lets an agent explore

The key to agent performance is less about making the LLM smarter, and more about designing a "field" where it can run hypothesis-testing loops efficiently — a slightly paradoxical idea: use the LLM without leaning on it entirely. Leave search, aggregation, computation and simulation to existing systems, and let the LLM focus on controlling the exploration — illustrated with a counterfactual shogi review system. (Japanese)

note · 2026.08.17READ ↗
 DEMO VIDEO FROM THE POST "What if the bishop were here?" — a local AI agent moves the board to explain the best move YouTube ↗
06 — CONTACT

Let's talk

Your challenge doesn't have to be well-defined yet. Inquiries starting from "we don't know what AI can do for us" are welcome.

EMAIL ME
contact@togami.org
LINKEDIN ↗