Hong Kong · Applied AI

AI that reads the room,
not the person.

Agentan builds sensing intelligence from signals that are already in the building. Ordinary Wi-Fi and millimetre-wave radar become a quiet layer that understands what is happening in a space — without a single camera or wearable.

First product in the field

Agentan Guard detects when an older person living alone has fallen — through walls, in the bathroom, at three in the morning — and calls for help while the fall still matters.

01 / Sensing

Signals as sensors

We treat radio as an instrument. Channel state information, radar returns and edge inference turn a flat into a sensor without adding one to every room.

02 / Models

Models that travel

Every home has a different floor plan. Our work centres on models that survive the move from lab to living room and keep their accuracy there.

03 / Delivery

Products, not demos

Hardware packaging, installation, alert routing and the dashboard the care team actually opens. We ship the whole path or none of it.

Products

One product in build, three behind it.

Everything Agentan makes comes from the same sensing core. Guard is where it meets a problem worth solving first.

In development · pilot 2026

Agentan Guard

A fall is rarely what kills an older person living alone. Lying undiscovered for hours is. Guard removes the two things that make discovery fail — remembering to press a button, and letting a camera into the bathroom.

  • Wi-Fi CSI from the router already in the home tracks movement through walls and around furniture.
  • A millimetre-wave radar covers the highest-risk room, usually the bathroom, with precision.
  • Inference runs on a plug-in edge box. Raw signal stays in the home; only events leave it.
  • An alert reaches family, the estate warden or the NGO's response desk within seconds.
WI-FI CSI · WHOLE FLAT RADAR · BATHROOM
Nothing to wearNo pendant, no watch, no charging. It works when the person forgets it exists.
No blind spotsRadio diffracts around walls and furniture, so the corner behind the sofa is covered too.
No imagesThere is no lens and no image to leak. The bathroom stays a bathroom.

Agentan Gait

The same signal that catches a fall also carries walking speed and stride symmetry. Gait watches those trends and flags a rising fall risk weeks before the fall.

Research

Agentan Risk API

Anonymised activity and risk signals for insurers and care operators, delivered as an API — for underwriting, claims prevention and care allocation.

Planned

Applied AI services

Model development, edge deployment and product design for teams with a data problem and a deadline. This is how the company funds its own R&D.

Available now
Technology

Two sensors, because each one fails where the other works.

Single-sensor fall detection is either blind or twitchy. Fusing Wi-Fi CSI with millimetre-wave radar is not about adding coverage — it is about each sensor vetoing the other's mistakes.

Radar alone

Sees a line, not a home

Millimetre-wave needs line of sight. A fall behind the sofa, around a doorway or in the next room simply does not register, and the alert never fires.

Wi-Fi covers it

Wi-Fi wavelengths bend around walls and furniture. The whole flat becomes one continuous sensing volume, including the parts nothing can see.

Wi-Fi alone

Sees motion, not always meaning

A whole-flat signal is broad. In the small, hard, high-risk room where most falls happen, it is harder to pin down exactly what moved and how fast it went down.

Radar covers it

One radar unit in the bathroom measures descent precisely — and because Wi-Fi responds to the water in a human body, a spinning metal fan stops reading as a person.

Research

Standing on a decade of published work.

Wireless sensing did not start with us. Guard is an engineering answer to a line of peer-reviewed results, ordered here the way they were built on each other.

2013 See through walls with Wi-FiShowed that Wi-Fi-band radio can pass through a solid wall and track a person moving behind it. Adib & Katabi · MIT · SIGCOMM
2017 WiFall: device-free fall detection by wireless networksEstablished that Wi-Fi channel state information alone is enough to identify a human fall. Wang, Wu & Ni · USTC / HKUST · IEEE TMC
2017 RT-Fall: real-time contactless fall detection with commodity Wi-FiMoved the algorithm off the workstation and onto ordinary off-the-shelf hardware. Wang et al. · IEEE TMC
2023 Rethinking fall detection with Wi-FiExtracted body descent speed from the CSI matrix; the authors report false-alarm and miss rates under 5%. Yang, Zhang & Zhang · Tsinghua · IEEE TMC
2024 XFall: domain-adaptive Wi-Fi fall detection with cross-modal supervisionThe result Guard is built around — accuracy that holds up when the model moves to a home it has never seen. Chi, Zhang, Ding, Ma, Yang & Du · Tsinghua · IEEE JSAC
Deployment

Three ways it reaches a living room.

Nobody shops for a fall detector. Guard travels to the home inside a relationship the household already trusts.

Route one

Social service partners

Pilot installations with Hong Kong NGOs and elderly-care organisations, funded through gerontechnology grants. Their fieldworkers already hold the keys and the trust.

Route two

Broadband bundles

Guard rides on the router. Offered as a line item when a family renews home broadband, an adult child can add it to a parent's flat in one step.

Route three

Insurers

A value-added service on senior health policies. Earlier discovery means shorter hospital stays, which makes prevention worth more to the insurer than to anyone else.

200,000Older people living alone in Hong Kong — the market Guard is built for.
50Homes targeted in the first year of pilots, through social service partners.
30MOlder residents across the Greater Bay Area, the market beyond the first city.
Team

Built by people who have shipped and people who have published.

Francis Tsang

Product & deployment

MSc in Data Science and Artificial Intelligence. Former product manager and consultant. Runs product design, partner relationships and getting the system into real homes.

Arthur Leung

Engineering

MSc in Computer Science. Former data analyst and app engineer. Owns the edge software, backend and the systems that keep alerts arriving.

Dr Chu Chun-fai

Assistant Professor, Computer Science, Hang Seng University of Hong Kong. Deep neural networks and human-motion prediction in medical engineering.

Dr Wong Wai-kit

PhD in Computer Science, University of Hong Kong. EAI Senior Member, Class of 2020. Databases, AI systems architecture and AI product development.

Dr Tsang Yau-keung

PhD in Chinese Medicine. Clinical analysis, and the work of turning sensor output into evidence a clinician will accept.

Contact

Every fall found in time is a night someone doesn't spend on the floor.

We are looking for NGO and care partners for the 2026 pilot, and for teams who want the sensing stack applied to their own problem.

Email tptfrancisdigital@gmail.com Based in Hong Kong SAR
Agentan — AI-driven services and products Guard is in development. Figures shown are targets, not performance claims.