Carson Kwok

About

AI & Automation Engineering | Enterprise Infrastructure

Senior Systems Engineer with 6+ years in enterprise IT, after four years supporting SFC-regulated trading floors. I build the AI agents and automation for a group of over 700 office users in eight countries; the largest puts about 65% of roughly 200,000 B2B sales orders a year into SAP with no re-keying. I also deliver business systems end to end in React and TypeScript, and remain design authority for identity, security and infrastructure.

Download CV (PDF)Classic site

Projects

AI Agents

HR AI AgentBuilt as a Microsoft Copilot Studio agent living inside Teams; it reads an HR policy knowledge base with natural-language understanding and writes approved leave back to the HR system — no coding for the end user.
Appraisal Data AgentA Copilot Studio agent that searches Microsoft 365 (Outlook, Teams, OneDrive) with natural-language retrieval, then calls the SharePoint API to assemble the scattered evidence into a structured appraisal draft.
IT Help-Desk AgentA Copilot Studio help-desk agent grounded on the IT knowledge base I maintain. It answers common L1 questions such as password resets, Wi-Fi and VPN, and walks staff through the self-service steps instead of changing accounts itself; anything it cannot settle goes to the service desk with the conversation attached.
Competitor & Market Intelligence AgentBuilt on Microsoft Copilot Studio: a scheduled agent that searches the web for competitor and new-product news each day, has an LLM assess the impact and opportunity, then posts the briefing to Teams and emails it to management, with no one running it.
Marketplace & Social Listening AgentBuilt in Microsoft Copilot Studio with an Agent Flow that calls the Apify API. Staff paste a marketplace URL or product ID in Teams and get back the product details, a summary, and customer reviews, comments and ratings from 12+ platforms, including Lazada, Amazon, Shopee, Tmall, Taobao, Instagram, Threads, X, YouTube, TikTok, Facebook, RedNote, Bilibili, Zhihu and Weibo.
Voice of Customer Report AgentA Copilot Studio agent that reads the reviews, comments and ratings collected by the Marketplace & Social Listening Agent and writes a senior-management or board-level Voice of Customer report: an executive summary, what customers like, what they dislike and how to improve. It compares our products with competitors, across platforms and between products, and delivers the report as a PDF and in Teams or by email.
Accounts Receivable AgentBuilt as an Amazon Quick skill deployed to an agent, for the Finance team's accounts receivable: submit a client's payment and invoice, and it reconciles them against the ledger in one reply — flagging remaining debts, overpayments, and how much credit-term clients have repaid versus still owe.
Concur Expense & Purchase Request AgentsBuilt on Microsoft Copilot Studio with Copilot Cowork skills. The expense agent reads a trip’s receipts and builds a Concur-ready Excel for the user to check; once confirmed, it submits the claim into Concur automatically through EmailXpenC, an SAP Concur App Center add-on that imports Excel expense data. A second agent creates Concur purchase requests from quotations and other documents.
SAP AI AgentBuilt on Microsoft Copilot Studio with Power Automate handling the SAP calls, published to both Microsoft Teams and Microsoft 365 Copilot — with authentication in front of every query, so staff pull live SAP figures from a phone or tablet instead of opening a desktop SAP session.
CV Screening AgentBuilt on Microsoft Copilot Studio and published to Microsoft Teams and Microsoft 365 Copilot: it reads every incoming CV, scores it against the role's real requirements and filters out the clear non-fits — so the hiring team only spends time on candidates worth reading.
SharePoint Document AI AgentBuilt on Microsoft Copilot Studio with Power Automate, indexed over the SharePoint document library and published to Microsoft Teams and Microsoft 365 Copilot: ask in plain language and it returns the answer together with the source document, instead of making people guess which file holds it.

RPA Pipelines

Zero-Touch SAP Sales OrdersChains a SharePoint e-form, a Power Automate trigger, Nanonets OCR to read the PDF, and a UiPath robot that types the data into SAP's desktop GUI — turning a paper-style order into a live SAP order with no human keying.
Automated Invoicing (Concur)A UiPath robot with Document Understanding OCR reads recurring vendor bills, fills an expense report from the extracted fields, and submits it into SAP Concur — replacing manual monthly data entry.
Dormant Account SweepA Python script that runs every day across Active Directory, Microsoft 365, SAP, MFG Pro, IBM Cognos, Copilot and Power BI licensing, applies the least-privilege account and licence rules, emails a grace-period warning, then disables dormant accounts — one job serving licence management, audit, security and compliance.
Daily Market Rates DigestTwo scheduled Power Automate flows that pull open-source rate data every day — base interest-rate benchmarks (HIBOR, SOFR, SORA, LPR) and currency exchange rates — build a comparison table, then auto-update a Teams channel and email the digest to the relevant users.

Full-Stack Applications

Marathon Sports POS — Shop Management SystemReact 18 + TypeScript on Vite with Material UI, React Router and React Hook Form. Eleven typed service classes (orders, products, product types, stock, shops, stores, users, dashboard) share one axios client over a generic base service. The API is a Python 3.11 Flask app packaged as an AWS Lambda function through an AWS SAM template with SQLite; a second TypeScript port runs on Netlify Functions over Turso (libSQL) for the cloud build, and a parity suite holds both to identical responses.
Marathon Inventory — Stock Tracking SystemReact 18 + TypeScript on Vite, styled with Tailwind CSS v4 and Radix UI primitives. TanStack Query holds server state, TanStack Table and Virtual keep long stock lists fast, and React Hook Form with Zod validates every form. ZXing drives the camera scanner and qrcode generates printable labels. The API is TypeScript on Netlify Functions over Turso (libSQL), with its own sessions, role and per-location scoping, audit log, rate limiting, password policy, mailer and schema migrations.

Web Platforms

GO Board Game PlatformBuilt with hand-written HTML and CSS (no framework), a custom calligraphy @font-face, and the lightweight canvas-nest.js library for the animated particle background.
Travel & Discovery SiteHand-built with HTML and CSS Grid, a glassmorphism style layer, an embedded Google Map, and vanilla JavaScript for the booking and login flows — no backend required.
300ADD Reflective JournalBuilt and published on Google Sites, which generates the pages without hand-coding — used here as a structured academic documentation portal.

Power BI

Brand Stores Sales PerformanceBuilt and published in Power BI Desktop, using Power Query to clean and model the source data and DAX measures for the sales, location and seasonal calculations.
Fried Chicken Sales RecordsBuilt and published in Power BI Desktop, using Power Query for data cleanup and DAX measures for the profit-margin and sales-channel calculations.

HR AI Agent

AI Agents · Built on Copilot Studio

HR AI Agent

Lives in Microsoft Teams. Staff apply for annual, sick or compensation leave in plain language and the agent books it against their balance; they also ask HR policy questions and get answers from the policy knowledge base — over 50 questions a month that no longer reach the HR team.

What it does

Leave requests: annual, sick and compensation leave booked against the employee’s balance and written back to the HR system.
Policy answers: grounded on a curated HR policy knowledge base, so every answer traces to a policy document.
Inside Teams: staff use it where they already work, with no new app to learn.

Built as a Microsoft Copilot Studio agent living inside Teams; it reads an HR policy knowledge base with natural-language understanding and writes approved leave back to the HR system — no coding for the end user.

Tools

Copilot Studio
Teams
Policy DB

Appraisal Data Agent

AI Agents · Built on Copilot Studio

Appraisal Data Agent

Gathers a year of scattered work from Outlook, Teams, OneDrive and SharePoint into one consolidated summary for the appraisal cycle, so staff write their self-appraisal from evidence rather than memory.

What it does

Retrieval across Microsoft 365: searches Outlook, OneDrive and Teams with natural-language queries.
Consolidation: turns scattered items into one summary of the year’s work.
SharePoint mapping: lays the summary out on the appraisal template.

A Copilot Studio agent that searches Microsoft 365 (Outlook, Teams, OneDrive) with natural-language retrieval, then calls the SharePoint API to assemble the scattered evidence into a structured appraisal draft.

Tools

Copilot Studio
SharePoint API
NLP

IT Help-Desk Agent

AI Agents · Built on Copilot Studio

IT Help-Desk Agent

Answers common IT questions in Teams from the knowledge base I maintain and walks staff through self-service fixes such as password reset — deflecting roughly 20 to 40 repeat tickets a month before they reach the service desk.

What it does

First-line answers: password, Wi-Fi and VPN questions answered from the IT knowledge base.
Self-service guidance: walks staff through Self-Service Password Reset rather than changing accounts itself.
Hand-off: passes anything unresolved to the service desk with the conversation attached.

A Copilot Studio help-desk agent grounded on the IT knowledge base I maintain. It answers common L1 questions such as password resets, Wi-Fi and VPN, and walks staff through the self-service steps instead of changing accounts itself; anything it cannot settle goes to the service desk with the conversation attached.

Tools

Copilot Studio
Knowledge Base

Competitor & Market Intelligence Agent

AI Agents · Built on Copilot Studio

Competitor & Market Intelligence Agent

Runs every morning to search for competitor and new-product news, assess the impact and opportunity for the business, and deliver a compiled briefing to management in Teams and by email.

What it does

Daily schedule: runs unattended without manual triggering.
News & web search: scans for competitor moves and new-product launches.
AI impact analysis: assesses business risk and opportunity from each item.
Two channels: posts the briefing to Teams and emails it to management.

Built on Microsoft Copilot Studio: a scheduled agent that searches the web for competitor and new-product news each day, has an LLM assess the impact and opportunity, then posts the briefing to Teams and emails it to management, with no one running it.

Tools

Copilot Studio
Teams

Marketplace & Social Listening Agent

AI Agents · Built on Copilot Studio

Marketplace & Social Listening Agent

Paste a marketplace URL or product ID in Teams and get the product information, a summary and what customers are saying — reviews, comments and ratings — from 12+ international and Chinese marketplaces and social platforms. Run several times a month to gather 30+ products for each report to senior management.

What it does

12+ platforms: marketplaces such as Lazada, Amazon, Shopee, Tmall and Taobao, and social media such as Instagram, Threads, X, YouTube, TikTok, Facebook, RedNote, Bilibili, Zhihu and Weibo.
One input: a product URL or ID is enough; the Agent Flow handles the Apify calls.
Feeds the Voice of Customer agent: the collected reviews are the input for board-level reports.
In regular use: several runs a month, 30+ products per report.

Built in Microsoft Copilot Studio with an Agent Flow that calls the Apify API. Staff paste a marketplace URL or product ID in Teams and get back the product details, a summary, and customer reviews, comments and ratings from 12+ platforms, including Lazada, Amazon, Shopee, Tmall, Taobao, Instagram, Threads, X, YouTube, TikTok, Facebook, RedNote, Bilibili, Zhihu and Weibo.

Tools

Copilot Studio
Agent Flow
Apify API
Teams

Voice of Customer Report Agent

AI Agents · Built on Copilot Studio

Voice of Customer Report Agent

Turns collected customer reviews into a board-level Voice of Customer report — executive summary, positives, negatives and how to improve — comparing our products with competitors, across platforms and between products. Delivered as a PDF and in Teams or by email.

What it does

Board-ready structure: executive summary, positives, negatives and recommended improvements.
Three comparisons: our products vs competitors, platform vs platform, product vs product.
One pipeline: built on the data from the Marketplace & Social Listening Agent.
Delivery: a PDF report plus a Teams message or email.

A Copilot Studio agent that reads the reviews, comments and ratings collected by the Marketplace & Social Listening Agent and writes a senior-management or board-level Voice of Customer report: an executive summary, what customers like, what they dislike and how to improve. It compares our products with competitors, across platforms and between products, and delivers the report as a PDF and in Teams or by email.

Tools

Copilot Studio
Teams
PDF

Accounts Receivable Agent

AI Agents · Built on Amazon Quick

Accounts Receivable Agent

Supports Finance on accounts receivable: when a client pays, the user submits the invoice and the client's account balance, and the agent returns the account status — whether other debts remain, whether the client overpaid, and for credit-term clients, how much has been repaid and how much is still outstanding.

What it does

Payment reconciliation: matches each incoming payment and invoice against the client's ledger.
Outstanding-balance check: lists any other invoices the client still owes.
Overpayment detection: flags amounts paid in excess and suggests credit-note handling.
Credit tracking: for credit-term clients, shows how much has been repaid and what remains outstanding.

Built as an Amazon Quick skill deployed to an agent, for the Finance team's accounts receivable: submit a client's payment and invoice, and it reconciles them against the ledger in one reply — flagging remaining debts, overpayments, and how much credit-term clients have repaid versus still owe.

Tools

Amazon Quick
AI Agent Skill

Concur Expense & Purchase Request Agents

AI Agents · Built on Copilot Studio

Concur Expense & Purchase Request Agents

Used mainly by senior management and their secretaries. A business trip brings dozens of receipts, which used to be typed into Excel and then re-keyed line by line into Concur. The agent reads the receipts, produces a Concur-ready Excel for review and, once confirmed, submits it to Concur automatically. A companion agent raises purchase requests straight from quotations.

What it does

Receipt intake: transport, meals, flights, hotels and more — dozens per trip.
Review before submit: the user confirms the Excel before anything reaches Concur.
Automatic submission: EmailXpenC imports the confirmed Excel into Concur, ending line-by-line re-keying.
Purchase requests: a second agent builds Concur purchase requests from quotations and documents.
In regular use: several times a month, saving the secretaries a great deal of re-keying.

Built on Microsoft Copilot Studio with Copilot Cowork skills. The expense agent reads a trip’s receipts and builds a Concur-ready Excel for the user to check; once confirmed, it submits the claim into Concur automatically through EmailXpenC, an SAP Concur App Center add-on that imports Excel expense data. A second agent creates Concur purchase requests from quotations and other documents.

Tools

Copilot Studio
Copilot Cowork Skills
Excel
EmailXpenC
Concur

SAP AI Agent

AI Agents · Built on Copilot Studio

SAP AI Agent

Puts live SAP data inside Microsoft Teams. Staff ask in plain language from a phone or tablet and get real-time figures back — no desktop SAP session needed — with authentication enforced before anything is returned.

What it does

Real-time SAP data: queries live records rather than a stale export.
Teams-native: works anywhere Teams does — phone, tablet or desktop.
Authenticated access: identity is verified before any SAP data is returned.
No desk required: removes the need to sit at a workstation just to check SAP.

Built on Microsoft Copilot Studio with Power Automate handling the SAP calls, published to both Microsoft Teams and Microsoft 365 Copilot — with authentication in front of every query, so staff pull live SAP figures from a phone or tablet instead of opening a desktop SAP session.

Tools

Copilot Studio
Power Automate
SAP
Teams
M365 Copilot

CV Screening Agent

AI Agents · Built on Copilot Studio

CV Screening Agent

For teams facing high volumes of applications: the agent reads and scores every CV against the role's real requirements, filtering out clear non-fits so your people only spend time on genuine candidates.

What it does

Reads and scores CVs against your job requirements.
Auto-filters applicants who clearly don't fit.
Built for high-volume hiring.
Consistent criteria: every application is measured the same way.

Built on Microsoft Copilot Studio and published to Microsoft Teams and Microsoft 365 Copilot: it reads every incoming CV, scores it against the role's real requirements and filters out the clear non-fits — so the hiring team only spends time on candidates worth reading.

Tools

Copilot Studio
Teams
M365 Copilot

SharePoint Document AI Agent

AI Agents · Built on Copilot Studio

SharePoint Document AI Agent

Ask a question in Teams and get the answer straight from the SharePoint document library — with the source file attached. Removes the hunt through folders when the library is large and nobody is sure which document holds the information.

What it does

Natural-language search across the entire SharePoint site.
Returns the answer and the source document together.
Teams-integrated — no need to open SharePoint and browse.
Built for large libraries where users don't know which file holds the answer.

Built on Microsoft Copilot Studio with Power Automate, indexed over the SharePoint document library and published to Microsoft Teams and Microsoft 365 Copilot: ask in plain language and it returns the answer together with the source document, instead of making people guess which file holds it.

Tools

Copilot Studio
Power Automate
SharePoint
Teams
M365 Copilot

Zero-Touch SAP Sales Orders

RPA Pipelines

Zero-Touch SAP Sales Orders

A sales rep drops an order and its PDF into a SharePoint e-form; Power Automate triggers the flow, Nanonets OCR extracts the order fields, UiPath keys them into the SAP GUI, and SAP returns a confirmed order ID with no manual data entry. Around 65% of the group’s roughly 200,000 B2B orders a year now run this way, with customers onboarded by order count as the IT Global Director set.

What it does

SharePoint e-form intake: reps submit the order plus its PDF attachment.
Power Automate trigger: detects the new item and starts the flow.
Nanonets OCR: reads and extracts the structured order data from the PDF.
UiPath into SAP GUI: keys the data in and captures the returned confirmation order ID.

Chains a SharePoint e-form, a Power Automate trigger, Nanonets OCR to read the PDF, and a UiPath robot that types the data into SAP's desktop GUI — turning a paper-style order into a live SAP order with no human keying.

Tools

SharePoint
Power Automate
Nanonets OCR
UiPath RPA
SAP

Automated Invoicing (Concur)

RPA Pipelines

Automated Invoicing (Concur)

Recurring monthly HR and Admin bills — utilities, office rent, staff shuttle buses, phone, courier and executive company cars, around 20 a month — are read by UiPath Document Understanding (vendor, amount, line items), filled into an expense report and submitted to Concur for approval. It has since been extended to other departments.

What it does

Recurring HR and Admin bills: utilities, rent, shuttle buses, phone, courier and company cars, around 20 a month.
UiPath Document Understanding: extracts vendor, amount and line items.
Fills the extracted data into an expense report.
Submits the report to Concur for approval; since extended to other departments.

A UiPath robot with Document Understanding OCR reads recurring vendor bills, fills an expense report from the extracted fields, and submits it into SAP Concur — replacing manual monthly data entry.

Tools

UiPath RPA
Document Understanding
Concur

Dormant Account Sweep

RPA Pipelines

Dormant Account Sweep

A daily Python sweep across seven systems — Active Directory, Microsoft 365, SAP, MFG Pro, IBM Cognos, Copilot and Power BI licensing — emails a grace-period warning, then disables genuinely dormant accounts and reclaims unused licences. The first run reclaimed around 100; now only the 10 to 20 that build up each day are left to catch.

What it does

Seven systems, one sweep: AD, Microsoft 365, SAP, MFG Pro, IBM Cognos, Copilot and Power BI.
Licence reclamation: frees unused Copilot and Power BI seats for reallocation.
Grace period: emails the owner before any account is disabled.

A Python script that runs every day across Active Directory, Microsoft 365, SAP, MFG Pro, IBM Cognos, Copilot and Power BI licensing, applies the least-privilege account and licence rules, emails a grace-period warning, then disables dormant accounts — one job serving licence management, audit, security and compliance.

Tools

Python
Active Directory
Microsoft 365
SAP
MFG Pro
IBM Cognos
Copilot
Power BI

Daily Market Rates Digest

RPA Pipelines

Daily Market Rates Digest

Two daily Power Automate flows that pull open-source rate data, then auto-update a Teams channel and email a comparison report to the relevant users: (a) base interest-rate benchmarks — HIBOR, SOFR, SORA, LPR; and (b) currency exchange rates.

What it does

Interest-rate benchmarks: tracks HIBOR, SOFR, SORA and LPR daily.
Currency exchange rates: pulls the day's FX rates.
Comparison table: compiles the figures into a day-over-day digest.
Automated delivery: posts to a Teams channel and emails the report to the relevant users on schedule.

Two scheduled Power Automate flows that pull open-source rate data every day — base interest-rate benchmarks (HIBOR, SOFR, SORA, LPR) and currency exchange rates — build a comparison table, then auto-update a Teams channel and email the digest to the relevant users.

Tools

Power Automate
Teams
Email

Marathon Sports POS — Shop Management System

Full-Stack Applications

Marathon Sports POS — Shop Management System

A complete shop management system, built to the standard of software a chain would actually run — not a mock-up. Staff get a back office for takings, stock, branches, orders and accounts; customers get a self-service kiosk with pickup numbers.

What it does

Four screens behind one router: login, dashboard, shop kiosk and admin console, each gated by a role guard.
Back office at a glance: today's takings, best sellers, branch performance and low-stock alerts.
Customer kiosk: browse, add, see the total, send to the counter, collect by pickup number.
Front end — React 18 + TypeScript on Vite, Material UI with Emotion, React Router, React Hook Form and notistack for toasts.
Service layer — eleven typed classes over a generic base service and a single axios client, so every screen calls the API the same way.
API — Python 3.11 Flask with a nine-table model (users, shops, stores, products, product types, orders, order items, stock, dashboard), deployed as an AWS Lambda function from an AWS SAM template.
Cloud build — the same API ported to TypeScript on Netlify Functions with Turso (libSQL), because Netlify has no Python runtime and a serverless filesystem cannot keep a SQLite file alive.
A parity suite compares the two backends response by response, so either can serve the front end unchanged.

React 18 + TypeScript on Vite with Material UI, React Router and React Hook Form. Eleven typed service classes (orders, products, product types, stock, shops, stores, users, dashboard) share one axios client over a generic base service. The API is a Python 3.11 Flask app packaged as an AWS Lambda function through an AWS SAM template with SQLite; a second TypeScript port runs on Netlify Functions over Turso (libSQL) for the cloud build, and a parity suite holds both to identical responses.

Tools

React 18
TypeScript
Vite
Material UI
axios
Python 3.11 Flask
AWS Lambda
AWS SAM
Netlify Functions
Turso (libSQL)
Role-based access

Open live demo ↗

Marathon Inventory — Stock Tracking System

Full-Stack Applications

Marathon Inventory — Stock Tracking System

Tracks how much of every item you hold and exactly where it sits — which shop or storeroom. Every movement is recorded with who did it and when, and once written an entry cannot be changed.

What it does

Eight modules behind one router: catalogue, stock, scan, locations, accounts, permissions, profile and a public product page.
Every movement in, out or between locations lands in a write-once ledger stamped with the user and timestamp, so history cannot be altered after the fact.
Printable QR label per item; the camera scanner runs on ZXing and opens /p/<code>, the only screen in the system that needs no login.
Front end — React 18 + TypeScript on Vite, Tailwind CSS v4, Radix UI primitives and lucide icons.
TanStack Query for server state, TanStack Table and Virtual so thousands of stock rows scroll smoothly, React Hook Form + Zod validating every form against a shared schema.
API — TypeScript on Netlify Functions over Turso (libSQL): sessions, role-based access, per-location scoping, audit log, rate limiting, password policy, email invites, pagination, search and schema migrations.
Full account lifecycle: invite and accept, lockout after repeated failed logins, forgot / reset / change password.
Tested with Vitest and 15 Playwright end-to-end specs — including an axe-core accessibility pass — behind build gates for two type-check passes, lint, bundle size, a secret scan and a deploy check.
Bilingual EN / 中文 throughout, responsive from phone to desktop.

React 18 + TypeScript on Vite, styled with Tailwind CSS v4 and Radix UI primitives. TanStack Query holds server state, TanStack Table and Virtual keep long stock lists fast, and React Hook Form with Zod validates every form. ZXing drives the camera scanner and qrcode generates printable labels. The API is TypeScript on Netlify Functions over Turso (libSQL), with its own sessions, role and per-location scoping, audit log, rate limiting, password policy, mailer and schema migrations.

Tools

React 18
TypeScript
Vite
Tailwind CSS v4
Radix UI
TanStack Query
TanStack Table + Virtual
React Hook Form + Zod
ZXing
Netlify Functions
Turso (libSQL)
Vitest
Playwright + axe-core

Open live demo ↗

GO Board Game Platform

Web Platforms

GO Board Game Platform

Developed and deployed a full interactive website focused on the strategic GO Board Gaming community using modern web frameworks for structural layout and UX.

What it does

Interactive Navigation: Multi-page architecture detailing Go associations across Japan, Korea, China, and Taiwan.
Bilingual: every page switches between Traditional Chinese and English.
Data visualisation: Google Charts showing how tournament prize money is distributed.
Dynamic Backgrounds: Canvas-nest.js implementation for interactive particle nodes.

Built with hand-written HTML and CSS (no framework), a custom calligraphy @font-face, and the lightweight canvas-nest.js library for the animated particle background.

Tools

HTML
CSS
Google Charts API
Canvas-nest.js

Open in New Tab ↗

Travel & Discovery Site

Web Platforms

Travel & Discovery Site

A photo-led travel website for a fictional Japan tour company, with a clear path from browsing destinations to booking a trip.

What it does

Modern UI/UX: Implemented glassmorphism, CSS grid layouts, and automatic background image sliders.
Interactive Mapping: Embedded Google Maps for regional context and geographic discovery.
Account screens: front-end mock-ups of log-in, registration and password reset (no back end).
Destination Showcase: Curated visual grid presenting top destinations like Tokyo, Kyoto, Osaka, and Hokkaido.

Hand-built with HTML and CSS Grid, a glassmorphism style layer, an embedded Google Map, and vanilla JavaScript for the booking and login flows — no backend required.

Tools

HTML
CSS Grid
Glassmorphism
Google Maps API

Open in New Tab ↗

300ADD Reflective Journal

Web Platforms

300ADD Reflective Journal

Academic reflective journal portal developed during the Bachelor's degree program, documenting continuous learning and IT project milestones.

What it does

Academic Documentation: Structured tracking of university module progress and milestones.
Google Sites Framework: Deployed using Google's rapid site generation architecture.
Reflective analysis: logs of how each problem was approached and how each project was structured.

Built and published on Google Sites, which generates the pages without hand-coding — used here as a structured academic documentation portal.

Tools

Google Sites
Technical Writing
Project Management

Open live demo ↗

Brand Stores Sales Performance

Power BI

Brand Stores Sales Performance

In-depth Sales performance tracking of Market Space, Mooncake Love, and Toy Story Land tracking location variants and seasonal impacts.

What it does

Market Distribution: Understand brand distribution across HK to optimise store strategies.
Brand Performance: Assesses revenue to identify which brands are more popular among consumers.
Geographical Factors: Explores how shopping center locations affect sales performance.
Seasonal Analysis: Time-series charts tracking sales peaks for holiday forecasting.

Built and published in Power BI Desktop, using Power Query to clean and model the source data and DAX measures for the sales, location and seasonal calculations.

Tools

Power BI
Data Modeling
Time Series Analysis

Open live demo ↗Read the case study

Fried Chicken Sales Records

Power BI

Fried Chicken Sales Records

Extensive sales records and profit performance metrics for Fried Chicken products across delivery and physical channels.

What it does

Time Series Analysis: Identifies sales patterns within specific time frames.
Product Analysis: Compares volume and revenue to provide data support for pricing strategies.
Sales Channel Analytics: Evaluates platform efficiency (e.g. Food Panda, Open Rice).
Profit Analysis: Combines cost and sales data to calculate true product profit margins.

Built and published in Power BI Desktop, using Power Query for data cleanup and DAX measures for the profit-margin and sales-channel calculations.

Tools

Power BI
DAX
Channel Analytics

Open live demo ↗Read the case study

Experience

Apr 2026 - Present

Senior Systems Engineer

GP Batteries International Limited · AI & Automation | Infrastructure | RPA | SAP

Promoted in April 2026 to focus on AI agents and intelligent automation, as one of five engineers in the AI & Automation team the IT Global Director formed from the application and infrastructure teams. Every agent and pipeline in the group's portfolio is my own build. Frontline service delivery now sits with the IT Support function, which I set standards for and review, across the group — over 700 office users in eight countries: Hong Kong, China, Singapore, Malaysia, Vietnam, Japan, Poland and the USA.

AI Agents & Automation Delivery

Own the group's AI agent and automation programme end to end: discovery with the requesting department, design, build, security review, launch and tuning. Every conversational agent and automation pipeline in the portfolio is my own build.
Build conversational agents on Microsoft Copilot Studio and Amazon Quick, delivered where staff already work — Microsoft Teams and M365 Copilot — for HR, IT, Finance, Sales, Procurement and senior management.
Build unattended automation on UiPath, Power Automate, Copilot Studio agent flows and Python, connected to SAP, SharePoint, Concur, Exchange and external APIs such as Apify, so agents act on requests rather than only answer them.
Choose the platform for each case — Copilot Studio, Amazon Quick, UiPath or Power Automate — weighing where the data sits, licence cost, volume and the maintenance the result will carry.
Deliver the SAP sales-order automation against the sequencing the IT Global Director set — onboarding customers by order count rather than order value — now covering around 65% of roughly 200,000 B2B orders a year.
Build the document intake layer on Nanonets OCR and UiPath Document Understanding, and keep each agent grounded on curated knowledge sources so every answer traces to a document.
Build every pipeline to fail safely — exception paths, retries and a defined manual fallback — and monitor production runs through failure alerts in email and Microsoft Teams.
Scope each agent's permissions to the minimum data it needs, using the identity and SharePoint permission model I own, so no agent can surface a document its user could not open.

AI Governance & Adoption

Work with departmental key users to scope what each agent should and should not do, then drive adoption through user guides, training, demos and one-to-one follow-up, announcing each launch to Office-licensed staff.
Evaluate the enterprise AI tooling the group licenses — Microsoft Copilot and Copilot Cowork, Anthropic Claude, OpenAI ChatGPT, Google Workspace and HeyGen — and write the comparison for the IT Director's decision.
Contribute daily to the AI & Automation team the IT Global Director formed — five engineers meeting every working day: report progress, help colleagues through blockers, review their designs on security, permissions and integration, and propose new processes to automate.
Reconcile licence allocation against actual usage, feeding the findings into the daily dormant-account sweep.

Technical Governance & Service Oversight

Set the technical standards and escalation thresholds IT Support works to, and chair the service review over 200+ Jira Service Management tickets a month — quality, SLA and what gets automated, documented or escalated. Frontline support for around 300 Hong Kong users sits with IT Support and the outsourced L1 desk.
Act as design authority for changes to group systems, approving proposals on risk, rollback plan and audit trail before anything reaches production.
Remain the escalation point for incidents the support tier cannot resolve, and lead post-incident reviews so the fix lands in the knowledge base or in an automation.
Own the group IT knowledge base I built from the ground up, reviewing every article before publication, and mentor IT Support and the L1 desk; coordinate SOPs and rollouts with in-house IT teams in China, Singapore, Malaysia and Vietnam.
Keep hands-on delivery for boardroom and corporate-event AV and Microsoft Teams meetings, and oversee asset and service tooling on Lansweeper, Fuji Xerox managed print, Exclaimer and Uniqode.

Identity, Security & Infrastructure Governance

Own the access model: least-privilege standards, approval routes and role definitions across Active Directory, Microsoft 365 and core business systems, with privileged access approved through Delinea PAM.
Maintain the audit-ready control chain in which every account, mailbox, VPN and access change carries written manager approval in Jira, and audit joiner-mover-leaver by sampling.
Own endpoint and email security policy on Palo Alto Cortex XDR and Cisco Secure Email, including the USB device-control exemption workflow, the SSPR and MFA estate, and phishing containment through Barracuda Sentinel.
Own the design of remote access on Entra ID SSO with per-service security groups, and the change-control standard for FortiGate firewall work.
Govern Veeam backup — Backup & Replication and Data Cloud, with off-site copies on AWS S3 — and VMware, Nutanix and Horizon VDI, and act as system owner for SAP ERP, SAP Concur, IBM Cognos, IBM Notes, Shopify and DocuSign.

Jun 2024 - Apr 2026

Systems Engineer

GP Batteries International Limited · Infrastructure & Security | ITSM | Identity & Access

Multi-entity group behind GP Batteries (global energy solutions) — over 700 office users across eight countries: Hong Kong, China, Singapore, Malaysia, Vietnam, Japan, Poland and the USA.

Service Desk Leadership & ITSM Governance

Directed daily IT operations for approximately 300 Hong Kong users across multiple group entities, coordinating with in-house IT teams in China, Singapore, Malaysia and Vietnam to keep SOPs, system-access approval flows and project rollouts aligned across the group.
Supervised a three-person outsourced L1 service desk (ServiceOne, China) and one on-site support engineer; governed OTRS and Jira queues averaging 200+ tickets per month through weekly review meetings, driving SLA attainment and reducing ticket ageing.
Led the migration of IT ticketing from OTRS to Jira Service Management: designed request types and multi-level approval workflows, authored end-user and IT-side operating guides, and ran the global adoption campaign for staff and regional IT teams.

Identity, Access & Lifecycle Management

Administered Active Directory, Microsoft 365 and core business system accounts and licences; enforced least-privilege, approval-based access and ran regular dormant-account and licence reviews to control recurring spend.
Owned end-to-end joiner–mover–leaver processes across group entities: AD and Exchange provisioning, laptop and workstation build, desk-phone deployment with DID assignment, and application entitlement setup; on exit, mailbox conversion to shared, licence reclamation, sign-in blocking, AD disablement, asset recovery and user data backup.
Enforced an audit-ready control chain in which every account, mailbox, VPN and system-access change was raised as a Jira ticket carrying written line-manager or department-head approval, giving internal audit and compliance a complete documented trail.
Managed privileged accounts through Delinea PAM, alongside SharePoint, shared-drive and security-group permissions under documented approval controls.

Security Programmes

Deployed a group-wide USB device-control policy through Palo Alto Networks Cortex XDR — owning planning, phased rollout, global user communications and the design of a manager-approved exemption workflow — closing a significant data-exfiltration and malware vector without disrupting operations.
Rolled out self-service password reset (SSPR) and multi-factor authentication (MFA) company-wide, including global announcements, enrolment guidance and sustained follow-up support, materially reducing password-reset ticket volume and removing after-hours reset dependency.
Administered Cortex XDR across the endpoint fleet, covering application allow and deny controls, agent version upgrades and software installation monitoring.
Supported email security operations on Cisco Secure Email (CES/ESA): phishing and spam quarantine review with release and block decisions, plus post-delivery recall and mailbox clean-up through Barracuda Sentinel during phishing incidents.

Network, Backup & Infrastructure

Modernised remote access from FortiToken-based local firewall accounts to Entra ID SSO, building per-service Entra security groups (SAP, file server and others) so VPN entitlements were granted by approved business role rather than blanket access; authored FortiClient setup guides for IT and end users.
Maintained FortiGate firewalls with scheduled configuration backups and controlled patching, taking pre-change VM snapshots and backups as standard rollback protection.
Operated enterprise backup on Veeam Backup & Replication covering VMs, servers and file servers plus endpoint and SaaS backup; monitored jobs daily, executed restores including file-level recovery of user-deleted data, and replicated NAS data to AWS S3 for off-site retention.
Migrated endpoint, OneDrive and Exchange backup from Druva to Veeam Data Cloud in response to a vendor price increase — preserving full coverage, avoiding a substantial cost uplift and consolidating the estate onto a single backup platform.
Supported VMware, Nutanix and VMware Horizon VDI environments, including VDI provisioning and account setup for IBM Cognos users.
Administered access and day-to-day operations for SAP ERP, SAP Concur, IBM Cognos, IBM Notes, Shopify and DocuSign.

Documentation & End-User Services

Built the group IT knowledge base from the ground up: joiner–mover–leaver runbooks, AD, Exchange and system-account procedures, VPN setup, data restore and error-handling guides, and standardised user communication templates — reducing dependency on individual knowledge and shortening L1 onboarding and handover.
Delivered conference and AV support for boardroom meetings and corporate events, covering Microsoft Teams setup and testing, camera and audio configuration and meeting recording, including on-site coordination with hotel venue teams for off-site events.
Managed workstation provisioning, IT asset inventory through Lansweeper, managed print services via Fuji Xerox AMDP, Exclaimer email signature management and Uniqode digital business cards.

Jun 2023 - May 2024

IT Officer

Winbull Securities International (Hong Kong) Limited · Trading & Settlement Systems | Infrastructure & Security

Hong Kong brokerage trading Hong Kong and US equities.

Responsibilities

Deployed an Apache Guacamole jump host server standardising secure VPN tunneling and remote RDP access for Windows/MacOS endpoints.
Integrated the Homarr Server Application Dashboard for centralised internal service routing.
Rolled out M365 authentication end to end and administered the Lark suite.

Nov 2020 - Mar 2023

IT Support

Get Nice Securities Limited · Trading & Settlement Systems | Market Data & End-User Support

Hong Kong brokerage covering securities, futures, asset management and corporate finance.

Responsibilities

Contributed to the IT office relocation plan, so trading infrastructure came back up quickly after the move.
Architected and deployed a dedicated Windows Server framework specifically isolating the iAsia Trading System.

Jul 2017 - Aug 2017

IT Internship

Hospital Authority — IT & Health Informatics Division · Hardware Diagnostics | IT Asset Inventory

Statutory body managing all public hospitals and medical institutes in Hong Kong.

Responsibilities

Diagnosed recurring hardware faults and wrote up the analysis.
Recommended fixes for the performance and reliability problems found.

Certifications

Gemini Certified EducatorGoogle · Verified
Artificial Intelligence FundamentalsIBM · Verified
Cloud Computing FundamentalsIBM · Verified
Cybersecurity FundamentalsIBM · Verified
Information Technology FundamentalsIBM · Verified
Web App & Database Dev for BusinessCity University of Hong Kong · Verified
Cisco Certified Network Associate (CCNA)Cisco · In Progress
Certified Information Systems Auditor (CISA)ISACA · In Progress

Also: Hong Kong driving licence (Transport Department, HKSAR)

Contact

Ready to architect robust IT infrastructure or deploy intelligent AI automation for your organisation? Reach me through any of the options below.

Email: carsonkwok.hy@gmail.com

LinkedIn Download CV (PDF)CK Academy

Terminal

Type help and press Enter.

carson@os:~$

Network

The infrastructure I designed and look after. Select a node to see what it does.

Cloud & SaaS: Cloud productivity suite and off-site object storage — the destination for replicated backups and the identity source for SSO.
Identity · SSO + MFA: Directory and single sign-on with multi-factor and self-service password reset. Access is granted per business role, on written approval.
Email Security: Inbound mail filtering with quarantine review, plus post-delivery recall and mailbox clean-up during phishing incidents.
Firewall · VPN: Perimeter firewall terminating role-based VPN access. Config backed up on a schedule; snapshots taken before every patch.
Core Network: Core switching between the edge, the compute cluster and the office floor across multiple sites.
Hypervisor Cluster: Virtualisation clusters hosting business servers, with physical-to-virtual migrations and snapshot-based rollback.
File & App Servers: File servers and line-of-business applications — ERP, expense, reporting and e-signature — with permissions under approval control.
Endpoint Fleet: Managed laptops and workstations with endpoint detection, USB device control and application allow/deny policy.
Virtual Desktops: Virtual desktop pools serving the analytics platform, provisioned per user on request.
Backup & Recovery: Daily backup of servers, file shares, endpoints and mailboxes; jobs monitored every day, restores tested, copies replicated off-site.

Stocks

Prices for a few technology companies, from TradingView (may be delayed). Nothing is loaded until you press the button.

The prices could not be loaded. Check your connection and try again.

Nothing matches.