As makers, close to the people who use it.

We plan and build our own products and hear directly from the people who use them. What we learn after launch, we bring back to the projects we work on with you.

Screenshot of the Karuttona official website

Karuttona

An app that helps physicians and other healthcare professionals write clinical records. It transcribes the conversation during a consultation on the device, organizes it by speaker, and assembles a SOAP draft. Audio processing, transcription, and SOAP generation all happen entirely on the device. iOS and macOS versions are in development. A Chrome extension for online consultations, “Karuttona for Web,” is being prepared for release as a limited beta for selected users.

Our role: Planning / design / development (all in-house; in development)

Visit the official website ↗ (opens in a new tab)

Privacy policy for Karuttona for Web (Chrome extension) (Japanese)

For clinics: check your own website for free (Japanese; no sign-up, about 1 minute)

Clinical data is not sent to the cloud

Audio, transcripts, SOAP drafts, and speaker profiles are never sent to the authentication or billing servers. Clinical data stored on the device is encrypted with AES-256-GCM using a device-specific key. “Not sending it” is designed in as a premise, not offered as a feature.

A clear line around record-keeping support

It does not assist in diagnosing or ruling out conditions, suggest treatments, make prescribing decisions, or assess urgency. All output is a draft; the premise is that qualified physicians check it against the original audio and create the official record based on their own judgment. We decide up front, on the product side, what it will not do.

Built to work in practice

Real-time transcription, checking and correcting speakers, and notation support from a medical terminology dictionary. The screens and controls are designed with one priority above all: not interrupting the flow of the consultation.

Tech stack: Swift / SwiftUI / On-device speech recognition / AES-256-GCM encryption / StoreKit

Screenshot of the SHIBA3 official website

SHIBA3

A community app for logging lawn care and sharing it with fellow lawn enthusiasts across Japan. Along with a care log that records mowing, watering, and fertilizing in one tap, grass-variety tags, and finding fellow enthusiasts in your area, it includes an AI feature for asking about problems by taking a photo. Available on the App Store and Google Play.

Our role: Planning / design / development / app review / operations and growth (all in-house)

Visit the official website ↗ (opens in a new tab)

AI consultation that reads the condition from a photo

It turns vague concerns such as “it’s withering” or “the color has faded” into words, from a single photo. Rather than passing AI output straight through, we designed usage limits, disclaimers, and non-definitive wording on the product side.

Getting all the way through store review

When the app was sent back in review (a deficient purpose string), we traced the cause back to the Info.plist in the binary, removed the permission itself, and resubmitted. We also added automated tests so the same incident cannot happen again.

Running user acquisition ourselves

We built the article media in-house with Next.js + microCMS, reflecting updates immediately with ISR. Compliance with Japan’s rules on undisclosed advertising (stealth marketing) is built in at the implementation level — for example, a PR label is added automatically only to articles linked to products.

Tech stack: Expo / React Native / Supabase / Next.js (admin and media) / microCMS / Firebase Analytics

Screenshot of the Kirokune official website

Kirokune

An iOS app for recording, on the same day, things that have happened to you, such as harassment. It brings audio recordings, photos, short notes, and a timeline view together, building up a record on your device that lets you explain what happened later. A fully native implementation in Swift / SwiftUI.

Our role: Planning / design / development / operations (all in-house)

Visit the official website ↗ (opens in a new tab)

Designed so records stay with you

Note text, audio, names, and case names are not sent for analytics. Measurement uses only fixed event names and predefined categories. Even when “we can’t know without collecting it,” we never fill the gap by turning real data into analytics attributes — a principle enforced in code and tests.

Assuming you don’t want to be seen

A display switch that anticipates someone looking over your shoulder, storage that stays entirely on the device (a local vault), and quick recording from a widget. We worked backward from the real situation of “I don’t want anyone to see me using this” to build the UI.

Chat messages as evidence, too

It can read text from screenshots and import chat histories, bundling scattered evidence into a single timeline.

Tech stack: Swift / SwiftUI / WidgetKit / StoreKit / On-device processing / Cloudflare Pages + microCMS (website)

Let’s figure out the next step together.

No spec required. Tell us about the problem or the idea — we’ll take it from there.

Talk to us