Stop 01 of 10 · Opening1–2 min

PolyU · COMP3424 · Industry Sharing & Technology Showcase

From Sensors
to Reality.

What changes when AI can see, understand, simulate and act?

Benny (Yan) Liao · Industry Engineer · Founder, HSIMC · Architecture & Strategy Advisor, InnoSpire Monday · 5 October 2026 · AG206 14:15–15:00 · flexible 10 stops · ~25–30 min core + discussion / demo

Start the field note Download PPTX · 10 slides

What to say · opening

Good afternoon. I am not here to teach you the definition of a digital twin — you already have a course for that. I am here as an industry engineer to share what I am seeing when sensing, simulation, AI and physical systems start to connect.

I began from applied physics and measurement, then standards, machine vision, IoT and manufacturing. Today the question has changed. A sensor no longer only produces data for a dashboard. Its output can become context for an AI system that reasons, simulates and sometimes acts.

So for the next half hour, I want to stay with one question: when AI can see and understand Reality, what does an engineer need to make the next action real, safe and verifiable?

Opening question to the room: “When you hear the word sensor, where do you think the sensor system ends?” Do not answer it yet — return to it at Stop 03.

Stop 02 of 10 · Industry context2–3 min

External Industry Practitioner · Sharing & Technology Showcase

A practical bridge from COMP3424 — sensing, simulation, IoT and digital twins — to what engineers are now seeing in AI-enabled physical systems.

Founder

Hongkong & Shenzhen Instant Manufacturing Corporation Limited · HSIMC

Exploring distributed GenAI-native capability production infrastructure — how digital intelligence becomes locally actionable in physical Reality.

Architecture & Strategy Advisor

InnoSpire Technology Limited

Smart glasses, AI-assisted visual support, inclusive mobility, and the engineering path from sensing to real-world assistance.

Chief Digital Strategy Officer

Pannex

Product engineering, global supply chain, manufacturing, and the practical gap between a digital idea and a deliverable product.

Hong Kong & Overseas Development Advisor

Kingsware · 金智维

Digital employee and enterprise AI deployment — another view of how AI capability enters real organisations and workflows.

Hong Kong public innovation context · HKSAR Government public records

A real public setting behind the smart-glasses discussion.

On 13 April 2026, the HKSAR Transport Department reported on Smart Traffic Fund projects showcased at the Smart Hong Kong Pavilion during InnoEX. The official release described an AI computer-vision smart-glasses project assisting persons with visual impairment, and documented government, industry, academic and research participation in the same public setting.

Smart Hong Kong Pavilion group photo from the HKSAR Government press release

HKSAR Government · 13 Apr 2026

Smart Hong Kong Pavilion · InnoEX 2026

Official public photo and report on the Smart Traffic Fund showcase.

Open official HKSAR press release ↗

Smart-glasses demonstration from the HKSAR Government press release

HKSAR public record · InnoSpire named

Smart glasses for visual assistance

HKSAR Social Innovation and Entrepreneurship Development Fund records explicitly name InnoSpire Technology Limited in smart-glasses ventures and publish project funding and outcomes.

Open HKSAR SIE Fund record ↗

Public references document the project and ecosystem context. They do not imply HKSAR Government endorsement of HSIMC, the speaker, or this session.

What to say · why this industry context matters

Before I go into sensors, I want to make clear why I am here. This is an industry sharing session, not a teaching appointment. My work sits across several realities: HSIMC is where I am exploring how AI capability becomes physically actionable; InnoSpire works on smart glasses and visual assistance; Pannex keeps me close to products, manufacturing and global delivery; and Kingsware gives another view through enterprise digital employees.

The two images here are useful because they move the discussion out of a laboratory. The Hong Kong public record shows a real ecosystem where government, industry, universities and research teams meet, and where smart glasses are discussed as assistance for people with visual impairment.

I am not showing this as an endorsement. I am showing it because once a technology enters a real public environment, engineering is no longer only about whether the model works. Users, safety, identity, access, responsibility and evidence all become part of the system.

Point to the public photo and say: “This is the kind of place where a sensor stops being a component and becomes part of a human system.”

Stop 03 of 10 · The sensor shift2–3 min

The sensor is no longer the endpoint.

For decades, sensing meant turning a physical phenomenon into a signal. Now the signal can become context for AI — and that changes what the system can do next.

SENSE

UNDERSTAND

SIMULATE

DECIDE

ACT

EVIDENCE

What to say · where does the sensor end?

When I studied applied physics, a sensor was very easy to locate in the architecture. A physical phenomenon became an electrical signal, then we measured it, calibrated it and decided whether the result was reliable. Later, IoT extended the chain: the signal moved through a network, into a database, a dashboard or an application.

Today that same signal can become context for AI. A camera may not stop at pixels. It can become objects, relationships, motion, state and possible intent. A distance sensor may contribute to a safety decision. So the engineering boundary moves.

This is why I use this six-step chain: sense, understand, simulate, decide, act, evidence. Not every system needs all six, and AI should not automatically act. But once the system crosses from observation to action, you must know what is allowed, what actually happened and how you can prove it.

Return to the opening question: “So now, where does the sensor end?” Let students answer for 20–30 seconds.

Stop 04 of 10 · One engineer's timeline2–3 min

1989 → 2026: the tools changed. The engineering question stayed.

1989

Applied Physics

Measurement, physical signals and the discipline of observing reality.

1997

Standards

GB 6844-1997: repeatability, interoperability and shared engineering language.

2015

Machine Vision

Industry 4.0, machine vision and cyber-physical systems.

2016

IoT + Robotics

Connected sensing begins to shape service-robot capability.

2017

Connected Value

NB-IoT, blockchain, global transaction and infrastructure questions.

2026

GenAI + Reality

Sensors, digital twins, agents, authority and physical action.

How do we know what is happening in Reality — and what can safely happen next?

What to say · the technologies changed, the engineering question did not

This timeline is not here as a CV. I want to show you how quickly the labels change. In 1989 I was studying applied physics and measurement. In 1997 I worked on a national standard, where the problem was repeatability and interoperability. By 2015 I was talking about machine vision and Industry 4.0. Then IoT, robotics, NB-IoT and blockchain became the new language.

Now the language is GenAI, agents and physical AI. Each wave is important, but engineers should not be trapped by the fashionable word of the year. Underneath all of them is the same discipline: what is the real state of the physical world, how certain are we, what can change, who is allowed to change it, and what evidence remains afterwards?

If you remember that underlying question, you can survive many technology cycles. Tools will change during your career. Reality does not become less demanding.

Pause on 1997 and 2026. Contrast “shared standards for products” with “shared protocols and evidence for AI capability.”

Stop 05 of 10 · Digital twins after GenAI3 min

A twin can become operational context for an agent.

Digital Twin

What is happening?

Represent relevant state, relationships and constraints from the physical system.

Simulation

What might happen?

Explore possible outcomes before changing the physical system.

AI / Agent

What should happen next?

Reason over context, propose or execute bounded next actions.

Physical Reality

Can it actually happen?

Capability, access, power, calibration, identity, authority and evidence still matter.

A good answer is not yet an action. A possible action is not yet an authorized action.

What to say · digital twin becomes operational context

Your course already asks you to connect digital twins and simulation with IoT and AI. I want to add one industry question: what happens when the twin is no longer only for visualization or analysis, but becomes context that an AI agent can use?

I separate the questions. The digital twin asks: what is happening now? Simulation asks: what might happen if we change something? AI can ask: what should happen next? But physical engineering has one more question: can it actually happen here, with this equipment, this user, this authority and these constraints?

This is where many impressive demos fail. A model can recommend an action. That does not mean a machine is available, calibrated, powered, connected or permitted to execute it. And even if an action happens, we still need evidence of what actually occurred.

Ask: “If a digital twin says a valve should open, what information is still missing before the real valve opens?” Take two answers.

Stop 06 of 10 · Summer Internship 20263 min

Give students Reality, not only exercises.

The summer team worked across software, AI, payment, IoT, smart glasses, prototyping, evidence and live web runtimes. Browse their work only if it is useful to you.

5 student hubs

Summer Intern Team · 2026

One directory linking the five student spaces and their project evidence.

nicholas90.homelinked.tech/team/ ↗

The experiment

Can a student make one capability work end-to-end?

Idea → code / CAD → physical or digital result → live runtime → user → evidence → continuation.

What to say · give students Reality

This summer I worked with five students, including PolyU students. I did not want the internship to be a sequence of tutorials. I wanted them to touch the whole chain: software, AI, payment, IoT, smart glasses, CAD, 3D printing, physical assembly, live URLs and evidence.

The important point is not that every prototype became a product. The important point is that a student can discover where an idea fails when it meets Reality. A CAD file may look correct but not assemble. A web page may work but nobody can pay. A camera may capture an image but that does not prove the task was completed. A good prototype teaches you where the missing capability is.

The link on this page opens their actual work. You do not need to read it now. If it is useful after the session, look at one student and ask: what did this person make possible that was not possible before?

Do not tour all 53 links. Open only one example if the room wants a demo; otherwise keep the directory as evidence for later exploration.

Stop 07 of 10 · MCP & callable capability2–3 min

Naming is not capability.

A domain can declare where a capability should live. Engineering begins when what is behind the address is real, callable, bounded, authorized and observable.

MCP-ready namespace example

mcp.makehere.tech

Open the current status. The page itself states the boundary between an addressable namespace and an active MCP capability.

mcp.makehere.tech/ ↗

Engineering boundary

Address ≠ Capability ≠ Authority

A URL, DNS record or discoverable tool does not by itself prove that execution is available, permitted, correct or accepted.

What to say · naming is not capability

You will hear a lot about MCP and AI agents. I own and use many domains because naming and addressability matter when capabilities need to be discovered and connected. But this is a good place to be strict about engineering language.

If I create mcp.example.com, I have created an address and a design intention. I have not automatically created an MCP server. Even when a real protocol endpoint exists, that still does not mean the capability is available to everyone, or that an AI has authority to execute it.

So I separate three things: address, capability and authority. This sounds simple, but it becomes very important once AI can call tools. A production system needs to know what exists, what it can do, who may use it, under what limits, and what evidence is returned.

Open mcp.makehere.tech only if useful. Use its current status as the example: declaring a future interface is different from claiming an active capability.

Stop 08 of 10 · Before graduation3 min

Can you make one real thing work end-to-end?

You do not need to master every discipline. But you should be able to cross enough boundaries to turn knowledge into a working, inspectable capability.

Can you sense or capture a real condition?

Can you explain what the data means — and what it does not mean?

Can AI or software use the context safely?

Can another system call or continue your work?

Can you show evidence that the result actually happened?

What to say · nine months is enough to build evidence

Many of you will graduate in about nine months. I would not spend those nine months trying to become an expert in every new AI framework. The frameworks will change. Instead, choose one real problem and make one capability work from end to end.

It might be a camera that detects something meaningful, a sensor that feeds a simulation, a digital twin connected to a real device, or a small agent that can call a tool. But do not stop at the first successful screenshot. Ask whether another person can use it, whether the limits are clear, whether failure is visible, and whether you can show evidence of the result.

If you can graduate with one project where you can explain the sensor, data, model, software, physical constraint, user and evidence chain, your portfolio tells an employer much more than a list of technologies.

Give students 20 seconds to scan the five questions. Ask them to pick the one they are currently weakest at.

Stop 09 of 10 · Continue exploring1–2 min

Use what is useful. Ignore what is not.

HSIMC

Main entry

Distributed GenAI-native capability production infrastructure.

hsimc.tech ↗

Read

Publications

Longer reading on infrastructure, practice and participation.

hsimc.tech/publications/ ↗

Student evidence

Intern team

Five student spaces from the 2026 summer work.

nicholas90.homelinked.tech/team/ ↗

Stop 10 of 10 · Closing thought1–2 min

REALITY CAN →
I CAN ACT.

Your value in the AI era is not only what you know. It is what you can make possible in Reality.

What to say · closing

AI is making answers cheaper and faster. That does not make engineering less valuable. It makes the boundary between an answer and a real capability more important.

For me, “Reality can → I can act” means this: the physical world actually contains the capability, access, authority and evidence needed for an action to happen. Only then does an AI suggestion become something a person or system can really use.

If you graduate able to connect sensing, digital models and AI to one real outcome — and explain the limits clearly — you will already be working on a problem that many companies are only beginning to understand.

Finish with: “What part of this chain would you like to challenge or discuss?” Then move to Q&A rather than adding another conclusion.

From Sensorsto Reality.

External Industry Practitioner · Sharing & Technology Showcase

Hongkong & Shenzhen Instant Manufacturing Corporation Limited · HSIMC

InnoSpire Technology Limited

Pannex

Kingsware · 金智维

A real public setting behind the smart-glasses discussion.

Smart Hong Kong Pavilion · InnoEX 2026

Smart glasses for visual assistance

The sensor is no longer the endpoint.

1989 → 2026: the tools changed. The engineering question stayed.

Applied Physics

Standards

Machine Vision

IoT + Robotics

Connected Value

GenAI + Reality

A twin can become operational context for an agent.

What is happening?

What might happen?

What should happen next?

Can it actually happen?

Give students Reality, not only exercises.

Summer Intern Team · 2026

Can a student make one capability work end-to-end?

Naming is not capability.

mcp.makehere.tech

Address ≠ Capability ≠ Authority

Can you make one real thing work end-to-end?

Use what is useful. Ignore what is not.

Main entry

Publications

Intern team

REALITY CAN →I CAN ACT.

From Sensors
to Reality.

REALITY CAN →
I CAN ACT.