Personalized On-Device AI for Communication

AI that learns your language, and runs by your side.

EqualFrontiers researches and develops small AI models that personalize to each individual's language and communication style, and run on smartphones and tablets.

Overview of Personalized On-Device AI: from limited physical input such as a switch (Limited Input), a personalized model trained on the person's messages, vocabulary, notes, and style (Personalized Model) generates communication in their own voice on a smartphone (Your Voice). Key features: Fewer Inputs, Private On-Device, AAC / Predictive Input

Our goal is not simply an AI that generates "correct" text.

It is an AI that learns the words, expressions, writing style, and communicative context a person actually uses — an AI that delivers "words that sound like you" with minimal effort.

From predictive text input for people with severe physical disabilities to AAC (Augmentative and Alternative Communication), we are bringing technology that adapts to each person's own way of communicating into real life.

Personalizing AI, instead of generalizing it.

General-purpose generative AI is designed to work broadly for many people. But what communication requires is not necessarily "text that is natural on average."

The words you use with family.
The expressions you use at work.
The technical terms you rely on.
Your own turns of phrase and the rhythm of your sentences.

EqualFrontiers uses each individual's communication data to build small language models specialized for their purposes and their own expression.

Fitting the AI to the person. That is what we mean by Personalized AI.

Toward AI that lives inside your smartphone.

Personal conversations and writing are deeply private data.

That is why we pursue not only giant cloud-hosted AI, but also the potential of small AI models that run on the devices people already use — smartphones and tablets.

By compressing and optimizing models, we enable fast on-device inference for predictive input and communication support in everyday life.

Personalized × On-Device
This combination is our path to AI that is closer to you, and more deeply adapted to who you are.

Our first use case: assistive communication.

For people with severe physical disabilities, entering even a single character can take significant time and physical effort.

EqualFrontiers is developing technology that predicts what a person wants to say next from just a few characters and the surrounding context. This goes beyond simple word prediction:

Not "what sentence is generally likely to come next,"
but "what would this person say next."

That is our goal.

Combined with predictive input, AAC, switch access, eye-gaze input, and other interfaces, we aim to enable self-expression that sounds like the person — with as few operations as possible.

Built by the people who need it.

This research began with the communication challenges of Katsuki Ono, EqualFrontiers' representative, who himself lives with a severe physical disability.

Unable to move his limbs or speak, Ono operates a tablet with a single switch mounted at his head. Using the communication data he has accumulated over many years, he developed a personalized language model that predicts his own words. Through multiple generations of model development and evaluation, he is also working on running small language models on-device, on a smartphone.

This line of research has been accepted at ASSETS 2026, the premier international conference on accessibility and computing.

We do not see disabled people as mere "users of technology."

People with lived experience define the problems, build the technology, evaluate it, and deliver it to society.

We believe that this development process itself leads to better assistive technology.

Technology

Personalized Language Models

Language models that learn an individual's vocabulary, style, and expressive tendencies from their own writing and communication data.

Small Language Models

Fine-tuning and optimization that make small models highly practical by narrowing them to a specific purpose.

On-Device AI

Models that run on smartphones and tablets, for low-latency, privacy-conscious AI.

Assistive Communication

Applications such as predictive input and AAC that reduce physical input effort while preserving the user's own voice.

From Research to Real Life.

EqualFrontiers does not stop at papers and experiments — we aim to turn AI research into technology used in everyday life.

We are currently working on productizing communication-support technology built on personalized on-device AI, and on joint research and development with companies and research institutions.

Every body is different.
Every voice is different.
Every life is different.

So AI should be different for each person, too.

AI that learns your language.
AI that runs by your side.

Personalized On-Device AI by EqualFrontiers.

← Back to AI Services

Contact

For inquiries about productization, joint research and development, or media coverage of our personalized on-device AI, please use the form below. Briefly include your project area and timeline if possible.