Portfolio

Three AI Systems I Designed and Built.

I am ACLP certified and I spent 10 years doing hands on marketing for SMEs. Everything I teach comes from systems I built myself. Two of these are running with Singapore clients. The third is a working demo you can open and try right now.

ACLP Certified Adult Educator 10 Years SME Marketing Former SME Owner 3 Systems Built End to End
What I Have Built

Not Slides. Working Systems.

Each one started from a specific problem worth solving. For each, I have included why it was built that way, because the reasoning is the part that matters.

01

AnneeMatthew  |  Nursing Bra eCommerce  |  Conversational AI

AI Sales Concierge: a full bra fitting inside Instagram DM

A customer messages the brand on Instagram and gets taken through a complete personalised fitting. The concierge sends a short measuring video, asks where she is in her maternity stage, walks her through the measurements one step at a time with reference images, calculates her size, and then recommends specific products with direct links to buy.

Why Instagram DM
Fit is the biggest reason customers hesitate to buy intimates online and the biggest reason they send them back. It is also awkward to ask a person. DM is private, it is where these customers already are, and it is open at 11pm when they are actually shopping. A website chat widget would have missed all three.
The Hard Part
Every bra style has its own size chart, so a generic band and cup calculator gets it wrong often enough to cause returns. In the demo below, the same measured 34D becomes an order size S in one bra and an order size L in another. The concierge holds the chart for every product, so it recommends per bra rather than issuing one generic size.
0:00 / 3:25

Unedited screen recording of the live concierge. Watch from 2:00 for the size calculation and the product recommendation.

Built WithInstagram Messaging API, Claude API, product size chart ingestion

02

Storyworks Consulting  |  Working Demo, Open to Try  |  Backend Automation

Intake CRM: turns an inbox into a qualified, researched pipeline

Enquiry emails arrive at one address. Before anyone opens them, each one is written up as a structured record: who the sender is, what their company does, how urgent it is and why, whose move it is next, and the specific action to take. No inbox migration and nothing to install.

Why It Works This Way
A contact form takes a message and stops. The work that actually costs an owner their morning is the research, the judgment call on priority, and remembering whose turn it is. Every ranking carries its reasoning on the record, so a wrong call is one you can see and correct rather than a black box you learn to distrust.
The Hard Part
Scam enquiries look exactly like real ones. The record below flags a sender claiming a Ministry of Education budget from a personal gmail address, and explains the pattern it matched before any proposal goes out. Getting that judgment specific to one business, rather than generic keyword rules, was the real work.
Intake CRM dashboard showing a qualified lead with next step, and a suspected phishing enquiry flagged with reasoning Open the live demo  →

Built WithNext.js on Vercel, Supabase, Claude API

03

芽笼益仁堂 TCM, Geylang  |  Healthcare Retail  |  AI Content Production

AI Video Pipeline: agency quality short form without an agency

A short form video pipeline for a traditional Chinese medicine clinic, mixing AI generated clips with real footage shot in the shop. Every video is built around one specific complaint a patient walks in with, and carries bilingual on screen subtitles so it lands with both Chinese speaking and English speaking audiences in Singapore.

Why This Approach
A TCM clinic cannot afford an agency retainer for weekly short form, and stock footage does not show their actual shop or their actual treatments. Mixing AI generated clips with real in store footage keeps the output honest and the production cost low enough to publish consistently.
The Hard Part
Health content has to stay credible. Every video opens on a symptom a real patient recognises, such as leaning to one side when walking, or fingers going numb after long hours at a desk. The hook is the patient's own complaint, not the treatment name, which is what makes a clinical service watchable.
TCM fire dragon moxibustion short TCM customer testimonial short TCM back tuina massage short TCM foot reflexology short TCM short on uneven leg length TCM short on stiff neck TCM treatment short TCM treatment short
Watch the full set on YouTube  →

Built WithAI generated video clips, real in store footage, bilingual subtitle production

What This Adds Up To

The Same Question, Whether You Are Hiring Me or Buying From Me

Conversational AI That Sells

Assistants that hold a real buying conversation, ask the right qualifying questions, and recommend the right product. Built on the channel your customers already use.

Automation Behind the Scenes

Systems that do the research, the qualifying, and the follow up chasing before a person ever opens the message. Built end to end, from the data model to the interface.

Training That Comes From Building

ACLP certified and registrable for WSQ delivery. Every example I teach from is a system I designed and built myself, so the lesson includes the decisions and the trade offs, not just the result.

If You Want This Built, or You Want the Training, Message Me.

Whether you are an SME founder who wants a system like these in your business, or a training company looking for someone who can teach AI from real builds, the fastest start is a WhatsApp conversation.

Start a WhatsApp Consultation Back to case studies and results  →