Kan Ninomiya二宮 貫

Kan is a 20-year-old software engineer and product builder in Hiroshima, Japan. He builds AI agents, Japanese language and speech systems, and web and mobile products, then runs what he ships.

20 years old / Hiroshima, Japan / Representative Director and President / COO at FP16 Inc. / Software Engineer

Featured work

Selected achievements with public sources and verification dates.

Grew Lina to more than 10,000 friends on LINE and transferred the business

I developed Lina, an AI chatbot for LINE, grew its user base to over 10,000, and later transferred the business.

More than 10,000 friends on LINE / Business transfer / Developed in October 2022

Independent evaluation of Japanese reading estimation

I developed a Japanese Reading Model combining candidate reading lattices and context rerankers, and compared it against public benchmarks.

100% on 66 difficult-reading questions / 91.9% on 260 independent evaluation questions / The comparison model scored 61.9%

Evaluated Japanese RAG with 2,000 questions

I compared Japanese RAG embedding models using a benchmark of 6 configurations and 2,000 questions.

6 configurations / 2,000 questions / Embedding evaluation for Japanese RAG

Operating at roughly 5 million page views per month

I develop and operate a closed web service portfolio with approximately 5 million monthly page views.

Approximately 5 million monthly page views / Closed web service portfolio

View detailed career and achievements

Media and public work

Television, press releases, talks, and other public work with sources.

TV Asahi: Hayashi Osamu no Ima Shiritai Desho!

Provided AI voice-cloning and AI avatar-generation technology. Through FP16 Inc., I provided AI voice-cloning and AI avatar-generation technology used in the program.

2024-05-23 / television

PR TIMES / FP16 Inc.: Opened the waitlist and began testing PEOPLE!, a voice-based social app for Gen Z

Planning, development, and test operations. I planned and developed PEOPLE!, a voice-based social app for Gen Z built around short voice posts, and launched a public test.

2024-07-03 / press_release

View all media work

What I am working on

Currently, I focus on AI agents, Japanese and speech processing, small internet products, and a public context foundation that humans and AI can share.

Operating AI Agents in Production

I am turning multi-agent development, review, research, and monitoring into an operational system with human approval and observability.

AI agents / MCP / automation / observability

Japanese and Speech Processing Stack

I improve reading estimation, speech-recognition post-processing, and TTS preprocessing using evaluation data and real-world usage logs.

Japanese / STT / TTS / evaluation

Small Internet Products

I continuously test consumer and developer products with a clear use case, a short path to launch, and room to improve from real usage.

0 to 1 / consumer / developer tools / validation

Public Context for Humans and AI

I publish the same public profile data through HTML, Markdown, JSON, and MCP, creating a public layer that AI can read safely.

public context / MCP / llms.txt / structured data

View all current priorities

What I can do

I choose the technology needed and carry work from concept through operation.

AI Agent Design and Operations

Agent system combining tool usage, MCP, memory, RAG, workflows, monitoring, and human approval.

AI agents / MCP / RAG / LLM / tool use / automation

Web Product Development

Full-stack web service development from 0 to 1, including frontend, API, authentication, data, payments, GraphQL, and public infrastructure.

TypeScript / React / Next.js / Hono / NestJS / GraphQL / Stripe / API

Consumer Mobile Product Development

Planning, UI, implementation, App Store release, CI/CD, and feature flags for iOS and Android apps using Expo/React Native.

Expo / React Native / iOS / Android / App Store / feature flags

Practical AI and Machine Learning Systems

Development integrating RAG, real-time speech AI, image anomaly detection, fine-tuning, and GPU inference into web and business systems.

TensorFlow / PyTorch / fine-tuning / OpenAI Realtime API / GPU / RAG

Cloud and Operations Infrastructure

Lightweight architecture centered on Cloudflare Workers, CI/CD, logging, monitoring, and staged deployments.

Cloudflare Workers / D1 / R2 / Durable Objects / AWS / GitHub Actions

Japanese, Speech, and Document Processing

Processing for integrating Japanese into practical systems, including reading estimation, post-speech recognition processing, OCR, and document generation.

Japanese NLP / STT / TTS / OCR / document generation

Product Delivery

Proceeding with requirements gathering, UI/UX, implementation, deployment, automation, and initial user validation as a single workflow.

product design / 0 to 1 / prototyping / operations / growth

View the full profile

Selected projects

A selection of active products, open-source software, research, and experiments.

Giga Site Delivery

In operation

A service that quickly turns HTML or ZIP files into shareable websites.

Personal contributionIdeation / UI design / Frontend / Backend / Infrastructure / Operations and growth

OutcomeOperated a web service for instantly publishing HTML and ZIP files

Web publishing / Cloudflare / SaaS

TRAC — Turning Travel Memories into Stories

Available on iOS

An iPhone app that combines location data, short videos, maps, and timelines to preserve travel moments as stories.

Personal contributionIdeation / UI design / Frontend / Backend / Infrastructure / Operations and growth

OutcomeDesigned and developed using Expo / React Native, and published as an iPhone app on the App Store / Implemented experiences with location data, short videos, maps, timelines, and automatic VLOGs / Built mobile CI/CD using GitHub Actions / Handled App Store review and implemented feature-flag-based controls

iOS / travel / consumer app / Expo / React Native / location / VLOG

Tashika

Public Demo

An API that corrects spoken technical Japanese and transcripts using context and domain vocabulary.

Personal contributionIdeation / UI design / Frontend / Backend / AI & Model Design / Infrastructure / Operations and growth

OutcomeProvided a public demo of the Japanese technical transcription correction API

Japanese / STT / API / LLM

Japanese Reading Model

In Active Development

Model/API experiments for Japanese reading and pronunciation estimation and TTS preprocessing

Personal contributionIdeation / Backend / AI & Model Design / Infrastructure / Operations and growth

OutcomeAchieved 100% on the 66-question difficult reading test and 91.9% on the independent evaluation of 260 questions / Published the model, API, and evaluation results

Japanese NLP / reading / TTS / model

Voice2Agent

Public Demo

A demo that turns voice memos into actionable instructions for Hermes, Claude Code, Codex, and ChatGPT.

Personal contributionIdeation / UI design / Frontend / Backend / AI & Model Design / Infrastructure

OutcomeProvided a public demo of creating work instructions for AI agents from speech

voice / AI agents / prompting / workflow

x-mcp

Open Source / In Operation

A read-only MCP server built with Cloudflare Workers that provides search and retrieval of public X/Twitter data

Personal contributionIdeation / Backend / Infrastructure / Operations and growth

OutcomeOperated and open-sourced a read-only MCP server handling public X data

MCP / Cloudflare Workers / TypeScript / X

View all projects

How I work

  1. Making small, fast, and functional things
  2. Prioritizing reversible changes and learning from real-world usage
  3. Evaluating not only technical completeness but also usability and operability
  4. Choosing simpler architectures before increasing complexity and maintenance debt
  5. Build testing, diff review, and human approval into AI-agent outputs

Data for AI agents

The same reviewed public data is available as Markdown, JSON, llms.txt, and Remote MCP.