Lead lea LLC — an AI company working across research and implementation
Get in touch"Giving data an expression" — we find the potential asleep inside data and bring it to life in business.
Our strength is moving between research and implementation with the same hands. In joint research with the National Center of Neurology and Psychiatry (NCNP) we work daily with procedures that hold up to peer review — leakage control, correction for multiple comparisons, sensitivity analysis — and carry that same rigor straight into DX support and AI product development for companies.
Redesigning business processes and setting the right DX strategy
Planning, building and operating our own AI products
Supporting decisions with accurate forecasting models
Lead lea original concept
Wiring a nervous system into the organization.
Departments are neurons. The AI agents connecting them are synapses. Controlling and optimizing those connections is harness engineering. Lead lea proposes "Synaptic DX", a way of treating digital transformation not as the rollout of isolated functions but as the design of an organization's nervous system.
Both the concept and the name originate with Lead lea LLC (IP filing planned)
In joint research with a national research institute, we are building indicators that measure interaction in conversation.
Lead lea × NCNP joint research
In November 2025 we formally contracted a joint study with the National Center of Neurology and Psychiatry (NCNP). Working from the Corpus of Everyday Japanese Conversation (CEJC), we defined 19 interactional features across four categories, nine of them newly defined, and are validating them against Big Five personality scores estimated by four large language models. Each feature is computed from a transcript with no manual coding step, so a third party can recover the same value. Research lead: Gen Fukuhara, CTO.
The figures come from a manuscript that has not been peer reviewed (revision of 30 September 2026) and may change through submission and review.
See the research in detailThe measurement pipeline from the paper, packaged so a person can run it on their own conversation. Recording through visualization is fully automatic, and results are shown only on the user's own device. Research use only, not a medical service.
About the toolRidge regression, permutation testing over 5,000 iterations, bootstrap variance analysis over 500 iterations, sensitivity analysis, and confound control by entering sex and age as predictors. On top of that, negative control conditions designed to falsify our own result.
Method and limitationsA paper on the development of Turn Lens was accepted as a poster at the 13th Conference of the Japanese Society for Support of Adults with Developmental Disorders, Kochi (September 2026).
Conference detailsTurning buried data into business value, from investigation through implementation to operation.
An emotion-analysis service combined with real-world explainable AI. Multiple affect scores from LLMs and scoring APIs are merged with device sensor data to generate a fitting response. Developed independently of our CEJC-based academic research.
Learn more (Japanese)Automated write-ups of ocean conditions using Amazon Bedrock and marine data APIs, delivering real-time reports and cutting documentation cost by up to 80%.
AWS blog (Japanese)A red-teaming platform for generative AI. Adversarial prompt generation, response evaluation and remediation guidance run end to end, streamlining AI governance.
Learn more (Japanese)An AI governance automation platform covering bias testing, drift detection, prompt guardrails and multi-model LLM evaluation, so post-deployment risk is handled automatically.
Learn more (Japanese)An autonomous target-selection AI agent. It analyses user data from several angles on its own and, working with Kankyo AI®, extracts promising targets automatically.
Learn more (Japanese)A LINE mini app offering personalised lunar biorhythm tracking and mental self-care, deriving each user's rhythm from their own data and suggesting suitable care.
Open the app (Japanese)End-to-end support for AI adoption, from planning through implementation to operational tuning.
We pin down the problem precisely, then design AI that fits the goal.
We propose a strategy that holds feasibility and ROI together.
We build accurate models shaped to the goal and the industry.
Rollout without downtime, so the change lands smoothly on the floor.
Continuous monitoring and tuning to keep systems stable.
Results delivered in finance, healthcare, manufacturing and retail.
We follow AWS best practices to keep operations both secure and efficient.
Work beyond the screen. Experience built in the ocean and in the kitchen is part of what Lead lea brings.
The more on-screen work AI takes over, the harder it becomes to replace experience built with the body, on site. Our CTO Gen Fukuhara has surfed for 40 years, ran a French toast food truck for ten years, and follows a vegan diet. Automated surf reports with Amazon Bedrock, presented at AWS Summit Japan 2024, are where years of watching the ocean met AI implementation.
40 years on the water. One-to-one lessons in Shonan and Oiso covering how to read waves, land training on a skateboard, injury-resistant conditioning, and plant-based eating and recovery. Authorised HOTLINE wetsuit dealer for over 20 years.
Lesson details (Japanese)A French toast food truck opened in 2012 and served at more than 740 events over ten years, with 36 media features. Ingredients are sourced to your budget, headcount and menu, and cooked at the venue. A vegan option without egg or dairy is available.
Catering details (Japanese)Drawing on 40 years of surfing and over 20 years of self-care practice, a talk on building a body that is less prone to illness and injury, for corporate wellness programmes, sports organisations and schools.
Talk details (Japanese)Building on the French toast food truck our CTO Gen Fukuhara ran for ten years from 2012, we have started an on-site catering service with no fixed shop. Ingredients are sourced to the client's budget, headcount and menu, and cooked at the venue; a vegan option without egg or dairy is available. Offered from Shonan and Oiso, Kanagawa, alongside private surf lessons.
AWS granted us USD 1,000 in credits that can be used only for Kiro, the AI IDE. We use Kiro across our research and product development, and in July 2026 we released a Japanese fork of "AI-DLC Workflows v2", a Kiro IDE harness. We will use the credits for spec-driven implementation and verification.
About KiroOur patent application covering the core technology of Kankyo AI® — "Multimodal emotion-analysis AI system" — was published on 29 September 2026 as JP 2026-151292 A. It was filed on 13 March 2025 (JP Application No. 2025-040035) solely by Lead lea LLC, with Gen Fukuhara as inventor: 5 claims, 12 pages, FI classification G06F 16/90,100. The disclosure combines emotion scoring with device sensor data (acceleration, gyroscope, battery level) in a single workflow and switches expert roles dynamically. A request for examination was filed on 26 September 2026 and the application is now under examination. Publication does not grant a patent.
Published application on J-PlatPat (Japanese)"Development and case report of Turn Lens, a conversation visualization tool based on quantitative interaction indicators" (first presenter: Gen Fukuhara) was accepted, and presented as a poster at the Kochi conference in September 2026.
Conference details (Japanese)We translated the prose of "AI-DLC Workflows v2", the AI-driven development lifecycle framework AWS publishes under MIT-0 (a Kiro IDE harness), and released it on GitHub. Strings the deterministic engine matches mechanically — stage slugs, enum values, audit event names — were frozen before translating 156 Markdown files, and the doctor health check still reports 31 passed / 0 failed. This is an unofficial derivative, not an AWS release.
Launch of Akaoni RT, a red-teaming platform that audits generative AI security automatically.
Read the release (Japanese)We formally signed a joint research agreement with the National Center of Neurology and Psychiatry (NCNP). The topic is quantitative indicators of interactional features in Japanese conversation — basic research using the Corpus of Everyday Japanese Conversation (CEJC). Because the CEJC licence prohibits commercial use of the corpus and of data derived from it, this work is carried out independently of our commercial products.
Kankyo AI® was registered as a trademark by the Japan Patent Office (classes 9 and 42).
Registry entry (Japanese)Common questions about the validity of the research and how that rigor carries into practice.
Research and implementation are handled by the same team. Because we work daily with procedures written for peer review — leakage control, correction for multiple comparisons, sensitivity analysis, negative controls — we can eliminate at the design stage the usual cause of "accurate in the PoC, not reproducible in production". On the implementation side we presented a production case study using Amazon Bedrock at AWS Summit Japan 2024, and we have carried projects from PoC to production in manufacturing, finance and healthcare.
Yes, they are welcome. Definitions for the 19 interactional features, the evaluation procedure (ridge regression, permutation testing, bootstrap variance analysis) and reproducibility scripts are all prepared, and a question about a single feature definition or about the evaluation design is entirely fine. Please look over the research in detail and write to genfukuhara@leadlea.com.
Validity is limited, and the design assumes that. The Big Five scores used here are item-level means of all 120 IPIP-NEO-120 items scored by four LLMs (virtual Big Five), not self-report measures. We therefore report inter-rater agreement alongside them and include a negative control in which the text-to-speaker pairing is broken; under that control all five dimensions lose significance. A summary-only condition, by contrast, remained significant while internal consistency collapsed, which we read as the model leaning on surface quantities when the input is impoverished. Convergent validity against questionnaire measures remains future work.
No. It is a tool for research use, not a medical service, and it offers neither diagnosis nor advice. The aim is to give a person and their conversation partner something concrete to look at together, without blame. Results are shown only on the user's own device. Short conversations inflate the numbers, so we emphasise reading differences between conversations rather than absolute values.
Most results that look like an immediate win are artefacts of the evaluation design. When the same subject appears in both training and evaluation data, accuracy rises easily and then fails to reproduce in production. We rule out coincidence first, with split designs such as subject-wise splitting, permutation testing and sensitivity analysis. On that basis Synaptic DX starts with one department and one agent, verifies the first value within a few weeks, and expands only as each step is confirmed.
The research uses the Corpus of Everyday Japanese Conversation (CEJC), built and released by the National Institute for Japanese Language and Linguistics. Its licence prohibits commercial use of the corpus and of any data derived from it, and we use neither in any commercial product. The Turn Lens case study covered only conversations for which the participants gave consent. For client work we design operations around anonymisation, access restriction, encryption and audit logging, aligned with the customer's own security policy. The guardrails and audit logs in Synaptic DX harness engineering apply that same control to the operation of the AI agents themselves.
Because both come from the same person's experience. Alongside AI research and implementation (joint research with NCNP, a talk at AWS Summit Japan 2024, a patent application for Kankyo AI®), our CTO Gen Fukuhara has surfed for 40 years, ran a French toast food truck for ten years from 2012, and follows a vegan diet. Automated surf reports built with Amazon Bedrock are one place where the two meet. Surf lessons and catering are offered in Japanese, based in Shonan and Oiso, Kanagawa.
Start with an informal conversation. Let's explore what your data can do. Research collaboration and replication enquiries are equally welcome.