The Copilot link opens GitHub Copilot Chat and needs a GitHub login. To ask how this page is built, add the repository charlestaysg/charlestaysg.github.io to the chat.
work
Live pages built from public data, explainer cartoons, and songs from my own life. No login needed for any of them.
How these are made
Every video here starts with something real: a year I lived through, or a question I kept hearing at work. I write down what happened, shape the words with Claude, and check every line against what took place. AI tools sing the songs and draw the scenes. Then I pick the takes, cut and fix them until they say what I meant.
Type its UEN or name to see its public ACRA record (status, age, what it says it does), the sector it sits in, that sector's hiring and training figures from the Pulse and the Landscape, what the public record can't tell you, and questions worth asking. The lookup runs in your browser against data.gov.sg; nothing is stored. Unofficial; ACRA is the official source.
For SME owners14 sectors · MOM data monthly · 81 SCTPs
SME Workforce Pulse
Where hiring is tight, and whether training matches it. Pick a sector to see vacancies, retrenchments and the career-switch courses for it.
Pick a sector to see job vacancies and retrenchments, how many PMETs were let go against PMET roles still open, and the career-switch (SCTP) courses the polytechnics and ITE run for it, with links to the skills and courses to train for. Built from MOM open data and public course pages; no login needed. Unofficial; MOM is the official source.
For SME owners26,099 courses · 515 providers
Skills Training Landscape
The supply side of the Pulse: SkillsFuture-listed courses for each sector, next to the PMET roles still open.
For each sector, see how many SkillsFuture-listed courses exist next to the PMET roles still open, who runs them, how long and how costly they are, how many are about AI, and which classes start in the coming months. Built from MySkillsFuture open data; no login needed. Unofficial; MySkillsFuture is the official source.
For SME ownerslive · NEA WBGT, 30 stations
Live Workplace Conditions
For people who work outdoors: NEA’s heat-stress reading for the nearest station, turned into MOM’s rest and water guidance.
For people who work outdoors, in construction, delivery, cleaning, landscaping or security. NEA's heat-stress reading for the station nearest where you work, turned into MOM's rest and water guidance, with a one-tap WhatsApp message to your supervisor, water and rest-break reminders, a heat-illness check and today's heat chart. Settings stay on your phone. Unofficial; MOM and NEA are the official sources.
For SME ownersguide · 3 printable sheets
Show What You Can Do
Making skills easy to see and check, with one door each for SME owners, mid-career workers and fresh graduates.
Singapore's October 2026 Skills-First paper asks employers to look at what people can do when they hire, design jobs, move, train and promote staff. This guide turns that into three things per skill (claim, proof, witness), a few steps for each reader, a printable sheet for each, and a box that turns one skill into a CV line and a LinkedIn line. Nothing you type is saved.
For SME ownersguide · 4 steps
Not Showing as an SME?
If your company isn’t on record as an SME, staff training costs more. How to ask for a review, and why to do it before you enrol anyone.
SME status for training funding is set from government data, and a correction is not backdated. This page walks through the official SME Status Application in four steps: check you qualify, get the Corppass role, have your headcount and turnover ready, then apply and wait for Approved. Checked against the official guide dated September 2026.
For SME ownersguide · 4 pressures
Fewer Hands, No First Rung
Fewer young locals, older staff working longer, foreign hires as a top-up, and AI taking the tasks juniors learn on. What a small firm can do about each.
Built around four hiring pressures, each checked against MOM and Population in Brief 2026 on 25 Oct 2026: resident fertility at 0.87, the re-employment age at 69, levies and quotas on work passes, and AI doing more entry-level work. Each comes with steps that cost effort more than money. Levy and salary figures are left out because they change often; check MOM before you act.
For SME ownersguide · 6 checks
Before You Let an AI Agent In
Personal AI agents like Meta’s Muse can now read your email and chats and act in your name. Six checks before you link anything that belongs to the business.
Written the day after Muse launched in Singapore (28 October 2026), from Meta’s announcement and my own try-out with personal accounts. Covers which accounts to link, approvals, customer data, payments, a rule for staff, the off switch, and posting to a shop Page. Menu names are as they were on 29 October; no prices, because Meta hasn’t published Singapore ones.
Live hazelive · NEA hourly PM2.5 · US AQI
Singapore Haze, Hourly
One page instead of two: NEA’s 1-hour PM2.5 next to the US AQI for each region, with wind, rain, haze news and a GP finder.
NEA's 1-hour PM2.5 sits next to the US AQI for each region, with wind, the 2-hour rain forecast, haze news, A&E links and a GP finder by postal code. Unofficial; NEA is the official source.
Why three copies? The same page runs in three places, each for a reason.
GitHub Pages: the main copy. Free, static, and opens on almost any network.
Cloudflare Pages ↗: adds what a static page can't do: live A&E waiting times and a wider news search, fetched by a small server. Some office networks block pages.dev, so use the GitHub copy there.
Manus ↗: an experiment. I asked the Manus AI agent to rebuild the page from the public code, to see how close it gets on its own. It adds Manus's own visitor analytics, and it may go offline.
Live hazelive · DOE APIMS hourly · EN/BM
Johor Bahru Haze, Hourly
The Johor side of the haze: DOE Malaysia’s hourly reading for each station on a map, in English and Bahasa Melayu.
DOE Malaysia's hourly API for each Johor station on a map, with wind arrows, satellite fire hotspots, a 24-hour trend, health advice by band and haze news, in English and Bahasa Melayu. Unofficial; DOE's APIMS is the official source.
Explainer3 min 33 s · narrated · captions
How Generative AI Works
Auntie Mei asks a chatbot for a cake promotion, and Tok the robot shows her what happens inside. The first of the cartoons.
More about this
A short cartoon for small business owners. Auntie Mei runs a bakery, and Tok the robot shows her what happens when she asks a chatbot for a cake promotion: tokens, next-word guessing, training, attention, why it makes things up, and where a small business should start.
Made with Claude, Kokoro and Whisper, offline on a Mac, with no paid AI services. AI-generated voice and visuals.
Explainer4 min 51 s · narrated · captions on screen
Ben & Auntie Mei
Over kopi, Ben and Auntie Mei work through one question about work or learning per episode. Job Hugging: why Singapore workers are staying put, what it costs them, and when it's time to move.
Five-minute cartoons for working adults. Ben is a mid-career executive and Auntie Mei is the friend who has seen it all. Every number on screen comes from a published source, and each source is shown when it's used.
Sources: MOM Labour Market Reports 1Q and 2Q 2026; ManpowerGroup Global Talent Barometer 2026 (Singapore); Federal Reserve Bank of Atlanta Wage Growth Tracker; Wrzesniewski and Dutton (2001). Made with Claude, Kokoro and Whisper, offline on a Mac. AI-generated voice and visuals. Not career or financial advice.
Explainer82.5 s · 1080×1080 · original score
Models as Insider Risks
Satya Nadella’s essay in 82 seconds: treat AI models like new staff, given only the keys they need and always able to be switched off.
Made with Claude Code, HyperFrames and GarageBand.
Based on Satya Nadella, "Models as Insider Risks in the Super Intelligence Era", X Article, 10 Oct 2026. Independent summary; not affiliated with or endorsed by Microsoft.
Explainerinteractive · 6 stages · 4 workplaces
Learning on the Job
How a new hire goes from watching to working alone. Pick a kitchen, a service counter, a factory floor or a bike shop, and watch the work move from coach to learner.
Six stages from cognitive apprenticeship: modelling, coaching, scaffolding, articulation, reflection and exploration. Each stage shows what the coach does, what it looks like on the job, and when to move on. Below it, five tips for anyone who trains people at work.
Based on Collins, Brown and Newman (1987). Workplace examples are made up for illustration. Built with Claude Code.
Stay safe4 free checks · links, messages, numbers
Check Before You Click
A link, SMS or call that feels off? The free checks I point friends and family to, starting with ScamShield and the 1799 helpline, plus what to do if you already clicked.
Four free services: the 1799 ScamShield Helpline, the ScamShield app, Malwarebytes Scam Link Check and Scam Number Check. Plus tips on what not to paste into a checker, why a clean result isn’t a promise, and why caller ID can lie. I’m not affiliated with any of them.
Experimentxinyao-style song + storybook video · 3 min 19 s · 中文, English subtitles
长大就懂 (You’ll Understand When You’re Older)
A song made of the things Singapore parents used to say: be home before dark, don’t sit so close to the TV, off the fan, you think this is a hotel? Years later, you hear the same words coming out of your own mouth.
More about this
It started from a poster of things every parent said. I swapped in the lines Singapore parents say, then followed them from an HDB flat in the 1980s to leaving home, to phone calls that only ever asked 吃饭了没有, to coming home late and finding one light still on. The pictures are drawn like a children’s storybook.
Created by Charles Tay · Lyrics written with Claude · AI-generated music and pictures (ChatCut, Qwen-Image in ComfyUI) · Video: HyperFrames
Experimentsong + music video · 4 min 30 s · 中文, English subtitles · piano
有树的地方 (Where the Trees Are)
A song for Xia Jinhe (夏瑾禾), Miracle Xiaxia, who died in Shanghai on 27 October, aged 23. At 19 she signed her own consent form to lose her leg. She ran a street stall on a prosthetic leg, turned down crowdfunding, and asked to be buried where there are trees.
More about this
I read about her on 30 October and kept thinking about the empty chair beside her hospital bed. The song is sung by a stranger, the way thousands of strangers left her the same words: 夏夏，晚安. There is no picture of her and no voice pretending to be hers, only the things and places from her story.
Song and lyrics by Charles Tay · AI-generated music, voice and pictures (ChatCut, Google Flow, ComfyUI) · Video: HyperFrames
Experimentsong + music video · 4 min 41 s · watercolour + a hand-drawn meter
So Far, So Good
A song about moving from teaching into a role where the work never quite closes. One hand-drawn meter carries the story: its needle creeps from green to red, then eases back with a little help.
More about this
It came out of a mentoring conversation. In a classroom the day ended at five and you closed the door. In the new role new things land every week, the calendar looks clean from outside, and you can’t book an hour for the thinking. A date on paper didn’t feel like anything, so the singer draws the load as a meter. A mentor’s voice sings the bridge: it’s not yours alone to hold. The video ends where it began, on the same week, with one block finally ticked.
Song, lyrics and video by Charles Tay · AI-generated music, voices and pictures (ChatCut, Google Flow) · Video: HyperFrames
Experimentmy first docudrama · 7 min 39 s · EN, 中文 subtitles · watercolour + archive film
The Night the Lights Stayed On
On Chinese New Year’s eve, 93-year-old Ah Ma puts her granddaughter to bed and tells her a real war story: the first Japanese air raid on Singapore, 4 am, 8 December 1941.
More about this
My first docudrama. Ah Ma and her family are made up; the air raid is real. 17 bombers came while the street lights were still on, 61 people died, and some who phoned for help were told it was only practice. The watercolours show what Ah Ma remembers, and the black-and-white film is real footage from December 1941. Her last line is the reason I made it: “That Sunday, everybody also said, nobody can touch us. Same sky. Same lights. Don’t take it for granted.”
A film by Charles Tay · Story and script by Charles Tay, written with Claude · AI-generated voices, music and pictures · Archive film: Australian War Memorial, US National Archives
Experimentmy second docudrama · 9 min 11 s · Malay & English, EN subtitles · watercolour
The Song Across the Water
On the night before National Day, a grandmother tells her grandson why Singapore has tanks. At 3.07 pm on 10 March 1965, a bomb went off in MacDonald House, where she worked.
More about this
My second docudrama. Nenek and her family are made up; the bombing is real. Elizabeth Choo, Juliet Goh and Mohammed Yasin Kesit died and 33 people were hurt, in one of 37 bombings in Singapore during Konfrontasi. The bombers hoped Malays and Chinese would turn on each other. The bomb killed two Chinese women and a Malay man. The facts come from the 1965 Straits Times on NewspaperSG.
A film by Charles Tay · Story and script by Charles Tay, written with Claude · AI-generated voices and pictures · Music: “Rasa Sayang” (traditional)
Experiment3 min 40 s · duet lullaby · EN, 华语 bridge · watercolour
Hold My Hand 牵着我的手
The song from the film. Ah Ma sings her granddaughter to sleep, and her father sings back from the night of the first air raid in 1941.
More about this
A slow lullaby duet in three-four time. Ah Ma tells the story of the biscuit tin; Pa is the one who phoned for help and was told it was only practice. In the Mandarin bridge he comforts his little girl, and she passes the same words on to Mei. The pictures are the film’s watercolours. I sampled three pairs of voices before picking this one.
Song and lyrics by Charles Tay, written with Claude · AI-generated voices, music and pictures · Made with ChatCut and Google Flow
Experiment4 min 49 s · lullaby · Malay, EN subtitles · watercolour
Lampu Suluh (The Torch)
The song from the film. Nenek sings her grandson to sleep on the eve of National Day: the bomb at MacDonald House, and the ordinary people who kept watch with a torch before Singapore had tanks.
More about this
A slow keroncong lullaby in Malay. At 3.07 pm the music stops and only rain is heard, then it carries on with the three who never came home. The pictures are the film’s watercolours. I made three takes and kept the one where every Malay line came through clearly.
Song and lyrics by Charles Tay, written with Claude · AI-generated voice, music and pictures · Made with ChatCut and Google Flow
Experiment3 min 6 s · city-pop · English · meme slideshow
Reset Tomorrow
11.47 pm, the AI helper is locked out and the report is due at nine. Out come a pen and an old note pad.
More about this
A song about finding out you can still do the thinking yourself. The pictures move like the meme slideshows people share on X: polka dots, card flips, star stickers, a glitch exit. I built the motion myself instead of using a template, and lifted the vocal after the first mix came out too soft.
Song and lyrics by Charles Tay, written with Claude · AI-generated voice, music and pictures · Made with ChatCut, Google Flow and HyperFrames
Experiment2 min 52 s · narrated story · four voices · illustrated
The Day the AI Went Quiet
A fourteen-person signage firm loses every AI tool on tender day. A pencilled ring binder, and the estimator everyone had stopped asking, get them through.
More about this
A made-up story for small businesses that now run on AI. It ends with one question, does your team have a Plan B?, and five things to write on one page: where your templates and rates live, who can do it by hand, who is learning, who to call, and the first fifteen minutes.
Story by Charles Tay, written with Claude · AI-generated voices, music and pictures · Made with Fish Audio, ChatCut, Google Flow and HyperFrames
Experiment3 min 24 s · xinyao folk · Mandarin, EN subtitles · watercolour
铁道旁的约定 (The Promise by the Railway)
Two schoolboys by the railway in 1986 promise to meet there when they are old. Forty years later, a message: “是我啊，还记得铁道吗？”
More about this
A campus-folk song in the spirit of the xinyao songs I grew up with. The station closed in 2011 and the tracks became the Rail Corridor. The song ends with a smaller promise: same place, next year.
Song and lyrics by Charles Tay, written with Claude · AI-generated voices, music and pictures · Made with ChatCut, Google Flow and HyperFrames
Experiment2 min 50 s · ballad · EN · notebook sketches
Still Searching
A song for anyone still looking for work, sung by an AI voice that isn’t a singer yet and tries anyway. Her lyrics and pencil sketches fill a notebook as she sings.
More about this
A mother in her forties hands back her staff pass after twelve years. Rejection letters at the void deck, her mother leaving the light on, the uncles at Bedok Jetty who don’t catch every time but come back every morning. I tested four ways of making a speaking voice sing and kept the one that sounded like a person trying, with nothing tuned. A second voice keeps telling her to step up.
Song, lyrics and video by Charles Tay · AI-generated voices and pictures · Made with Fish Audio, Logic Pro, Google Flow and HyperFrames
Experiment4 min 8 s · folk-pop · EN, 中文 · watercolour
Iron Rice Bowl 铁饭碗
A man spends twenty years between the vendor and the users. Then a polite letter arrives, strangers online call his job jiak liao bee, and he has to start again at 52.
More about this
I wrote this after reading what people online say when office roles are cut: take the package, those jobs were never real work, a six-week course won’t turn anyone into a coder. Some of that is true, and the song admits it. What it adds is the part the threads skip. Someone who has spent years between vendors and users knows how projects fail, and a small firm with a system nobody understands needs exactly that. The verses are in English, and the chorus and bridge are in Mandarin with English underneath. The man, the factory and the forum are made up, and one enamel bowl follows him through every chapter.
Song and lyrics by Charles Tay · AI-generated music and visuals · Made with ChatCut, Google Flow and HyperFrames
Experiment3 min 9 s · pop jingle · EN
Pending Concurrence
A love song filed through the proper channels: a circular, an annex, a shortlist and three months’ probation.
More about this
I wanted to see how far a campaign jingle could go if it stayed completely serious while the romance went through paperwork. An applicant clears his inbox, gets a circular, applies before the window closes, and is told to bring a five-slide deck on his intentions. It ends with a text message and a reply that says “Noted with thanks.” Everything in it is made up, including the scheme, the app and the stamps. The pictures came from one character sheet so the faces stay the same, and every cut lands on the beat.
Song and video by Charles Tay · AI-generated music and visuals · Made with ChatCut, Google Flow and HyperFrames
Experiment2 min 50 s · Short · Singlish · hawker food
Don’t Knock It Till You Try Lah
A hawker-centre singalong about the strangest food combos in Singapore: kaya toast in laksa, satay sauce on chendol, chilli crab on ice cream.
More about this
It started with a comedian’s reel about gross food combos. I wrote new lyrics around local food instead, sang a rough draft with an AI model on my own laptop, then made the final song in ChatCut. The pictures come from one cast sheet of five cartoon characters, so the uncle, the auntie and the cousin look the same in every shot. Which one would you actually try?
Song and lyrics by Charles Tay · AI-generated music, voice and visuals · Made with Claude, ChatCut, Google Flow and HyperFrames
Experiment4 min 9 s · duet · EN, 中文 · guzheng
哪里还痛 (Where It Still Hurts)
A song for Karen Nie (聂), our friend and physiotherapist for over ten years, who died suddenly this year.
More about this
My wife, my friend Charlie and I saw Nie every month for ten years. It started with a haircut while I waited for my laptop to be fixed. She came from a village in China, trained as a physiotherapist here, and always opened with the same question: which part is sore today, shoulder, leg or back? The song turns that question around. The pictures are watercolours, and she is only ever seen from behind. A guzheng plays her out over the credits.
Song and lyrics by Charles Tay · AI-generated music, voice and visuals · Made with ChatCut, Logic Pro, Google Flow and HyperFrames
Experiment3 min 14 s · piano ballad · EN
Three Thousand Years (The Intern’s Lament)
A very serious ballad about an intern who signed for “three thousand” and never asked the unit.
More about this
A headline about year-long internships got me wondering what happens if the contract just says “three thousand”. The video is a museum tour: the same intern at the same desk, from a Pharaoh’s court to the end of the stars, with a small placard for each era. Along the way an AI joins as an intern and is the boss by Thursday, and the review form still says “needs more time”.
Song and video by Charles Tay · AI-generated music, voice and visuals · Made with ChatCut, Google Flow and HyperFrames
Experiment1 min 29 s · instrumental · GarageBand
Every Sound Came From a Mac
A beat made only from the alert sounds already on my Mac and iPhone. No instruments, no sample packs.
More about this
I saw a film where a producer builds a track out of the sounds Apple devices make, and wanted to try it with what was already on my laptop. The Glass alert became the chords, Bottle pitched down became the kick, the screenshot shutter became the snare, and Submarine became the bass. Each sound comes in one at a time, then they all play together, and it ends on a long chord. The stems sit in a 12-track GarageBand project. In the video, every flash and lane lands on the note it belongs to.
Music and video by Charles Tay · AI-assisted music and visuals · Made with GarageBand, Claude and HyperFrames
A song about my first trip to Beijing: two weeks in winter while I was between jobs.
More about this
My first time in Beijing, and my first real winter. The first week was visits to tech companies, seeing how far automation and AI have gone on the factory floor. I went with my teacher, someone familiar, and met a lot of new people. The second week I was on my own: a cold two-hour queue, record shops, and good food. It was a break from the worry of not having a job. I came home still unsure, but with a different picture of China.
Song and video by Charles Tay · AI-generated music, voice and visuals · Made with ChatCut, Google Flow and HyperFrames
My 2025 told through the 5,301 questions I asked an AI search engine that year. AI sang it. I lived it. Every line happened.
More about this
A new home, a retrenchment, classes again, solo trips up north, a camera, and a pair of headphones that brought music back into my days. This year I read every one of those questions again and turned the year into a song.
A song by Charles Tay. Words from my own 2025, shaped with Claude. Music and vocals AI-generated (ChatCut). Video made with HyperFrames.
Experiment3 min 44 s · sing-along · EN, 中文, BM, Singlish
Pull Up a Chair
A sing-along about six generations at work: see people, not labels, then slow down and listen.
More about this
Written for coaching sessions with middle managers. Seven characters, from a grandmother of 85 to a 14-year-old at open house, each pull up a chair until everyone sits at one table. The final chorus carries three habits: notice what's showing, ask what's behind it, change how you listen.
Concept and lyrics by Charles Tay, written with Claude. Music and vocals AI-generated (ChatCut). Illustrations AI-generated (Qwen-Image in ComfyUI). Made with HyperFrames.
Experiment3 min 23 s · sing-along · workplace learning
Pull Up a Chair (at work)
The workplace-learning version: everyone at the bench has something to learn and something to teach.
More about this
Same hook, new story. A retired founder, a senior technician, an operations manager, a team lead, a new hire on OJT and a student on a visit each pull up a stool until the whole team sits at one workbench.
Concept and lyrics by Charles Tay, written with Claude. Music and vocals AI-generated (ChatCut). Illustrations AI-generated (Qwen-Image in ComfyUI). Made with HyperFrames.
Experiment4 min 11 s · English and Malay
Same Sky, Different Name
A folk-pop song about the Sep–Oct 2026 haze, sung in English and Malay. It sits well next to the haze pages.
More about this
Singapore and its neighbours breathe the same smoke and wait for the same rain. The lyric video moves from a grey skyline and a climbing PSI reading to clear sky.
Lyrics by Charles Tay, with AI assistance. Music and vocals AI-generated with Mureka in ChatCut. Made with Claude and HyperFrames.
Experiment4 min 57 s · duet
NACE Feelings
A 70s-style duet about what it feels like when someone finally helps your team learn. It started as a colleague’s joke.
More about this
The boss sings first, the team answers, and by the last chorus everyone is singing together.
Lyrics by Charles Tay, from a colleague’s idea. Music and vocals AI-generated with Mureka in ChatCut. The watercolour scenes are AI video, made in ChatCut; the rest was made with Claude and HyperFrames.
Experiment75 s · 1920×1080 · 5 sections · sing-along lyrics
Sunny Side
A happy little song. The Chinese and Japanese versions are new songs, written to be sung, not translated line by line.
The sun, clouds and confetti move on the beat and the chord changes, and each lyric lights up as it's sung.
Music and video by Charles Tay. Made with Claude, MiniMax-Music3 and HyperFrames. The music and vocals are AI-generated.
Experiment3 min · Mandarin duet
家就是你 Home Is You
A duet about family that follows one day at home, from the kitchen at sunrise to the walk home under lanterns.
More about this
He takes the first verse, she takes the second, and they sing the choruses together. Each lyric lights up as it's sung, tagged for who is singing.
Lyrics by Charles Tay. Music and vocals AI-generated with Mureka in ChatCut. The watercolour scenes are AI video (Seedance, in ChatCut); the rest was made with Claude and HyperFrames.
Experiment3 min 51 s · 3D animation
Carve Her a Heart
A folk song with a 3D story: his wife’s last wish is to hold something he made, so he carves her a small wooden heart.
More about this
He works at his bench day and night. At dawn it's finished, with “always” cut into it, and the bench is empty.
Story and direction by Charles Tay. Lyrics written with Claude. Music and vocals AI-generated with ChatCut. 3D animation in Blender; edit and lyrics in HyperFrames; made with Claude Code.
Experiment3 min 58 s · lyric video
Not Waiting Alone
A duet for anyone in between jobs, or stuck waiting for a door to open. It says one thing: you are not forgotten, and you are not alone.
More about this
The words come from a year of quiet trips: three weeks in the highlands, then a train and a flight to two old churches. Each lyric is coloured for who is singing, amber for him, blue for her, white when they sing together.
Words by Charles Tay. Music and vocals AI-generated with Mureka in ChatCut. Lyric video built with Claude Code and HyperFrames. Backgrounds are Pixabay stock (Iconikmg, Nils_Vega). The candle is an AI-generated image (Higgsfield), animated in Google Flow.
Experiment4 min 7 s · lyric video
Light Left On
My teenage years in one song. Fantasy books and PC games kept me steady while both parents worked. My grandmother left the light on.
More about this
Pixel art for the years when the world was a screen in a dark room, paint for when the stories opened up. Each verse is a place and a year: a flat in the late 1980s, a 486 and a pile of copied disks, a bus into town, a guardhouse in National Service.
Words and story by Charles Tay. Music and vocals AI-generated with Mureka in ChatCut. Images made with Qwen-Image 2.1 on my own Mac; pixel art drawn in code; five short clips from Wan 2.7 in OpenArt. Video built with Claude Code and HyperFrames.
Experiment3 min 18 s · vertical · sing-along
Don't Play Play (With the Haze)
A Singapore song for the friends who say "I feel fine" and skip the mask. Fine doesn't mean the air is fine. One air cleaner at home, in the room where the family sleeps.
More about this
It started on 15 October 2026, when people posted photos of rainwater in pails turned grey after the rain. The song follows one HDB estate: the pail in the corridor, the uncle who never checks the PSI, Ah Ma who likes the window open, a child coughing in the next room. The tips at the end follow NEA and HealthHub: windows shut when the PSI is unhealthy, a HEPA air cleaner in the bedroom, a mask outdoors. Speak to a doctor about your own health.
Idea and direction by Charles Tay. Lyrics drafted with Claude. Music and vocals AI-generated in ChatCut. Lyric timing with Whisper, pictures from Qwen-Image 2.1 and short clips from MiniMax H3, all run on my own Mac. Video built from code with HyperFrames and Claude Code. The characters are illustrations, not real people.
Experimentinteractive · one HTML file
Sausage Dog Password
A password field where a dachshund stretches with every character you type. Long passwords are the hardest to guess, so keep going until he’s proud of you.
Strength is graded on a sausage scale, from cocktail sausage to the full bratwurst, and a ruler shows his length in cm. A capital letter perks his ears up, a number earns a collar, a symbol earns sunglasses. Type a common password and he falls asleep. Nothing you type leaves the page.
Concept inspired by designer Ramin Nasibov’s stretching-dachshund password animation. This is an independent take with original artwork and copy. Built with Claude Code: the dog is hand-drawn SVG, stretched by a small spring animation in plain JavaScript.
Experimentinteractive · sound on · 20 s
Resistance Is Useful
My AI notes system works like the Borg from Star Trek. Press play to receive the collective’s hail, then see the one thing the Borg get wrong about learning.
The hail is a parody written from scratch. One AI voice clip plays three times, slightly out of step, and your browser mixes it live into a collective. Then a human answers. Below that: where the Borg joke fits, where it breaks, the lunch test, and two habits for anyone who files what they learn.
Voices by Fish Audio (AI-generated), cube in three.js, built with Claude Code. The notes system idea comes from Andrej Karpathy’s LLM Wiki. Star Trek and the Borg belong to Paramount. This is a fan homage, not affiliated with or endorsed by Paramount.
Experimentinteractive · sound on · 35 s
I Will Check You
An AI phones you to brag about its very particular set of skills: summarising, guessing, sounding completely sure. Answer the call, then make the same speech for your own job.
A nod to the phone call in Taken, flipped: the AI makes the threat, and a human answers with “I will check you, I will test you, and I will find your mistakes.” Below the call: what the AI is good at and the check that goes with each skill, a speech maker for nine jobs, and the five-answer test. Nothing you type in the speech maker leaves the page.
Voices by Fish Audio (AI-generated), with the phone-line sound added live in your browser. Built with Claude Code. Written from scratch apart from one catchphrase; Taken (2008) is a EuropaCorp film, and this page is not affiliated with or endorsed by its makers.
Experimentone image
Same Number, Much Shorter
A friend sent me a “Hire me, I have Claude Max” poster. This is my reply. A subscription isn’t a skill: the twenty years are what make the Max 20x worth anything.
Two CVs side by side. The 2025 one lists the work; the 2026 one only lists the plan I pay for. Both have a 20 in them, which is the joke and also the point.
Drawn as a web page and saved as an image with Claude Code. No AI-generated pictures.
news
Latest headlines
Connecting
Singapore business, jobs and AI headlines from CNA, The Straits Times, The Business Times, Vulcan Post and BBC, refreshed hourly.
Loading headlines…
Headlines and links belong to their publishers. Not affiliated with any of them.
At an INSEAD forum in Singapore this week, researchers warned that new staff now ask AI the questions they used to ask a supervisor. Each person gets faster. The team talks less, and its ideas start to look the same. In a small firm, that conversation between an old hand and a new hire is most of the training, so it’s worth keeping on purpose.
Sir Patrick Stewart died on 28 October, aged 86. For many of us he was Captain Picard: calm, fair, ready to admit he was wrong, and still able to make the hard call.
I made a three-minute photo tribute, from Hamlet in 1980 to Picard in 2019, ending with lines from Sonnet 73. Thank you, Sir Patrick. Engage.
Two more tributes worth watching: Rest in Peace, Jean-Luc by Rikerish ↗, and a Picard tribute by Target Audience ↗
Photos from Wikimedia Commons, openly licensed and credited in the video description · AI-generated music (ElevenLabs) · Sonnet 73 by William Shakespeare · Rikerish’s and Target Audience’s videos belong to their makers.
In April, a Bee Cheng Hiang employee used a personal AI tool to write a script for a members’ mailing. The prompt asked for “mass email using a local list” in batches, but didn’t say each person should see only their own address. One bracket sat in the wrong place, so every batch of 1,000 went out with all the addresses in the To line. The test only checked the send logs, not the email itself.
95,364 members were affected. PDPC says the AI tool didn’t malfunction; it wrote what it was asked for. It is the first AI-related breach reported to PDPC, which accepted a voluntary undertaking in September.
A separate rule took effect on 20 July. If you use customers’ personal data to train or fine-tune a generative AI model, you need a notice written for that purpose. A line in your privacy policy isn’t enough, and you can’t refuse service to people who say no.
What a small business can do:
test
before any bulk send, mail a test to two of your own inboxes and open it.
second check
someone else reviews AI-written code or mail settings that touch customer data.
one page
write down which AI tools staff may use for work, and which data stays out of them.
vendors
if a supplier wants to train AI on your customer data, ask what notice your customers will get.
training
every organisation must name a Data Protection Officer. A course for that person: the Practitioner Certificate in Personal Data Protection. Find a class on MySkillsFuture ↗ (polytechnic runs are also on STEP ↗; search the course name).
Meta launched Muse here on 28 October. It is free for most uses and links to Gmail, Google Calendar, Instagram, Shopee, Lazada, Grab, Agoda and more. Unlike a chatbot, it sends messages and buys things for you, after asking. Useful for an owner’s own errands. For the business, go slowly:
Four hiring pressures, and what a small firm can do
Fewer young locals, older staff working longer, foreign hires as a top-up, and AI taking the tasks juniors used to learn on. Each has a fix that costs more effort than money:
pool
hire mid-career and train for the job.
seniors
write down what they know before they leave.
passes
plan on the headcount you can keep.
juniors
keep some routine work human, with a senior checking.
A post this week said the biggest barrier to AI in companies is messy knowledge and data. I half agree. Messy files stop AI being useful after the first week, but most SMEs haven’t reached that week yet. Skills and habits come first, and the tools are mostly ones you already pay for:
storage
the Google or Microsoft drive you already have. Don’t add a second one.
housekeeping
one naming rule and one owner per folder.
questions
NotebookLM or Copilot on top of those folders.
people
record your experts doing the job. Some of what they know never makes it into a file.
About 1 in 3 Singapore employers plan to add staff from October to December, and about 1 in 5 plan to cut. ManpowerGroup’s survey nets that out to +13%: the same as last quarter, and down from +20% a year ago. Job worries are at their highest since Blackbox Research started asking in 2023, with about 1 in 5 people naming jobs as a top concern.
Trade and logistics leads the hiring plans. Finance and insurance improved the most, and UOB says it will take in 1,000 fresh graduates over the next year. The information sector, which covers software and media, fell furthest.
Two things I’d keep in mind. The survey counts plans, not hires. And the slowdown didn’t start this quarter; it arrived by mid-year and has held. Resident unemployment is still around 3%. What has changed is how long it takes to get in: fresh graduates and people who lost a job are waiting months longer than before. The article puts much of the software slowdown down to AI. That’s the writer’s reading, and a fair one, but the survey doesn’t measure it.
UOB’s move is the part I’d point an SME owner to. A big employer decided entry-level jobs are still worth having when AI takes over the routine tasks. Someone still has to learn the work well enough to check what the AI produces.
What a small business can do:
hire now
good people are waiting longer for offers. If you’ve been putting off a hire, this quarter gives you more choice than last year did.
try first
a short paid task tells you more about a candidate than a polished CV.
keep the junior role
if AI now does the routine part, give the junior the checking. They learn the work by checking it.
offer learning
you may not match a bank’s pay. You can offer real work early and someone to learn it from.
Summarised on 24 Oct 2026 from Vulcan Post’s report on ManpowerGroup’s Q4 survey and Blackbox Research’s poll. My own view. Not affiliated with Vulcan Post or ManpowerGroup.
CNA followed workers in their 50s who lost their jobs. In April to June, about 4 in 10 of them found work within six months, down from about 5 in 10 the quarter before. Vacancies still outnumber jobseekers, so the jobs are there. Getting hired is the hard part.
The part I’d point an SME owner to: Singapore research found that older workers who learned on the job had better job outcomes than those who relied on formal courses. Contacts, and work other people have seen them do, count for more than certificates. One man in the story retrained, sent over 400 applications and got no interviews.
What a small business can do:
learn at work
give staff over 50 a real new task and someone to learn it from, rather than only sending them on a course.
change the job
routine work gives people no reason to learn. Add one part to the job that needs a new skill.
write it down
keep a short record of what each person can now do, so the skill is easy to check later.
hire across
an older worker from another industry can bring skills you need. The woman in the story moved from garments to medical-parts manufacturing through a career conversion programme and now works as a data analyst.
Two of the three people CNA spoke to got back in through someone they knew, not a job ad.
A hawker-food entrepreneur’s post this week asked how an unprofitable small firm can give raises, and why anyone would trust a CV in the age of AI. The second question has an answer: when CVs and interview answers can all be polished by AI, what a person can actually do is the only signal left worth checking. Short work trials and real tasks tell you more than a well-rehearsed answer.
The first question is harder, and I don’t think policy has explained it to small-business owners in their own terms yet.
Copied pages take photos and reels from public accounts, build a following, then use it to sell something or run a scam. Five things that make it harder:
mark
put a faint watermark across the middle, not in a corner where it gets cropped off.
family
keep photos of children off public accounts altogether.
settings
turn off downloads, reuse and embedding on Instagram; set Facebook posts to Friends.
look
run your profile photo through a reverse image search every few months.
report
use Meta’s copyright or impersonation forms, not the “Report post” button.
Written 22 Oct 2026. Meta moves its settings around; check the current app before you rely on a step. Not affiliated with Meta.
· worth knowing · online security
An ad that can hack a phone without a click
Researchers at Citizen Lab found spyware that reached Al Jazeera staff through online ads in 2024 and 2025. Nobody had to click: the ad only had to load on a web page or in an app. A company posing as an ad seller bought ad space and aimed the ads at chosen people by location and internet address. Once in, the spyware could copy passwords, documents and messages, and take screenshots. Google has since cut the ad company off its platform.
What a small business can do:
update
turn on automatic updates for phones and browsers. This attack needed a flaw in the browser or app, and updates close those flaws.
restart
restart phones once a week. Some spyware doesn’t survive a restart.
fewer ads
an ad blocker on work laptops means your devices open fewer ads.
the few at risk
staff who handle payments, contracts or personal data can turn on Lockdown Mode (iPhone) or Advanced Protection (Android).
Each batch of targets cost millions, so most SME staff won’t face this exact attack. Cheaper versions of the same trick, called malvertising, are common.
Summarised on 18 Oct 2026 from Citizen Lab’s report and Al Jazeera’s video. I’m not a security specialist. Not affiliated with or endorsed by Citizen Lab or Al Jazeera.
BizSG is a new government site that puts business support in one place. If you run an SME, here is what to try first:
grants
a grant directory that says it can pre-check which grants you may qualify for, using data the government already holds.
matching
describe what you need and an AI recommender suggests grants. You need to log in, and for now it covers Enterprise Singapore’s EDGE grants and SWDA grants.
overseas
tariffs, free trade agreement savings, import rules and market data, for when you sell abroad.
I’ve only looked at the public pages, not the logged-in tools, so try the recommender yourself before relying on it. Grant terms change; check the agency’s own page for the latest.
Anthropic has about 3,500 staff and may list on the stock market next month. Its CEO says up to 40% of his time goes to culture, not to building AI. Why it’s worth an SME owner’s time:
time
once the company grew, he stopped making every call and spent his time keeping people pointed the same way.
write it down
every two weeks he reads staff a three-to-four-page document he wrote, then takes questions.
tell the truth
staff hear the real problems in plain words, and can ask him anything on an open channel.
trust
he hires people he trusts, so he can hand over decisions.
Tech in Asia sat down with Li Hongyi, director at Open Government Products, on how useful things get built inside a bureaucracy. If you’re trying to change how a large organisation works, this is what you’ll get out of it:
success
a way to think about it when the product never ships.
learning
why taking a month off work to learn “hidden” information can pay off.
building
how things actually get built inside a bureaucracy, and what the public sector can learn from high-performing private companies.
culture
his view that it can’t simply be built.
Tech in Asia says the full story and Q&A are on their website.
Interview by Tech in Asia. Topics are taken from their post; I haven’t added to them. Not affiliated with or endorsed by Tech in Asia or Open Government Products.
· worth watching · CNA · 9 min
Start with the problem, not the AI
Singapore is rolling out three new AI programmes for smaller businesses, with up to 12,000 places. CNA spoke to Johnson Poh, Assistant Chief Executive at IMDA, about what SMEs should do with them. Nine minutes, and most of it is advice an owner can use on Monday.
Why it’s worth an SME owner’s time:
start
with what eats your team’s time. Sometimes the fix is simpler than AI and works just as well.
data first
scattered documents and spreadsheets are the usual blocker. One of the programmes (with Singtel) turns them into data AI tools can use.
try first
IMDA keeps a list of pre-approved solutions you can test before committing, and wants most of them AI-ready within two years.
skills too
the programmes come with training and help to put tools in place, not just access to them.
the test
is the business better off? Time saved, customers served faster, costs down, revenue up. Using AI because everyone else does isn’t a reason.
Nearly one in four Singapore SMEs now use AI, up from one in seven a year earlier. If you’re not one of them yet, this is a calm place to start.
Video by CNA, 13 Oct 2026. Summary is my own reading of the interview; not affiliated with or endorsed by CNA or IMDA.
· video · 33 s · 9:16
The alarm nobody hears
Three years of AI alarms. “AI will take your job.” “Entry-level is dead.” “This changes everything.”
Janan Ganesh wrote in the Financial Times about “pessimism fatigue”: when warnings keep not coming true, people stop listening, and then miss the one that matters. AI talk at work has the same problem. Doom and hype wear people out the same way.
The warnings that still get heard are specific:
what
invoice matching
who
your industry
when
the next 12 months
watch for
a competitor quoting in hours, not days
You can check a warning like that. If it’s wrong, you’ll know, and so will I.
Made with GarageBand (original score), Claude Code and HyperFrames.
After Janan Ganesh’s column on pessimism fatigue, Financial Times. Independent commentary; not affiliated with or endorsed by the Financial Times.
about
I take something new and make it work for people who didn’t ask for it.
That has been my job for over twenty years. The subject kept changing: 3G phones for dealers, every iPhone launch for a call centre, 3D laser scanners for police and air-crash investigators, then designing courses and keeping them up to standard at colleges. The habit started on my own time: twenty years of trying tools.
now
Most business owners I meet don’t need another course for their staff. They need their people to handle new work without the boss stepping in. That’s what my team and I help with: training on the job, and putting new tools to work, AI included. I don’t write code, but I build with AI: the live dashboards and explainer videos on this page are made that way. Outside work, I write songs about my own life and let AI sing them.
if you run a business
We start with one question: where does your team’s time go?
We pick the one job that eats the most of it.
We fix it with the simplest tool that works, AI or not.
We train your people on their real work until they can do it without us.
Some problems need more than training. For those, I bring in specialists who know the trade, in supply chain, food and agriculture, hospitality and tourism, sports and health, and robotics and drones. You deal with one person, and I find the right expert. If your team still needs us a month later, we haven’t done our job.
Not sure where to start? Message me on LinkedIn. There’s often government support for this kind of work, more than most owners realise. How I work →
Focus
Workplace learning design
Training for working adults: on the job, in class and online.
Focus
AI adoption at work
Careful, tool by tool, starting from the work and not from the hype.
Focus
AI governance, explained plainly
Helping non-technical leaders see why AI needs rules it can’t rewrite for itself.
Focus
Visual explainers
Short animated videos that make one idea clear.
2026–Helping Singapore companies get their people ready for new work.With a team of specialists, starting from where the time goes.
2025–26Went back to class.AI automation, then photography and video with AI. Best homework I’ve ever done.
2022–25Checked everything before the auditors did.Kept courses up to standard, so audits held no surprises.
2020–22Got courses approved for SkillsFuture funding.And won training-partner status with PMI and IMDA.
2019–20Moved staff training online.The year the whole world had to.
2017–19Found out what was really going wrong, then built the training to fix it.For corporate clients, who often asked for one thing and needed another.
2016–17Tried new training small before rolling it out.Mapped what staff needed to learn next, then ran small pilots first.
2014–16Trained call-centre teams.Where you find out fast if a lesson worked: the next call tells you.
2012–14Taught investigators to capture a scene in 3D.Police, forensic and air-crash teams. Still my most unusual classroom.
2009–11Got a call centre ready for every iPhone launch.Staff had to know the phone before the queues formed.
2004–08Trained dealers for a nationwide 3G launch.And call-centre teams on mobile tech, with a short detour into insurance sales. My first training job, and I never moved out.
2000–04Fixed computers and answered the questions people were too shy to ask.Explaining tech to people who don’t love tech turned into a career.
1999Started in HR paperwork.Where it all began.
The full story, including the parts I left out, is on LinkedIn ↗.
ACTA · IAL 2014
SCTP Automation with AI · 2025
SCTP Photography & Videography with AI · 2025
Adv. Cert. Gen AI Apps Design & Prompt Engineering · SMU Academy 2026
BCom, HRM & Industrial Relations · Curtin
toolkit
The AI tools I actually use, and what for. I try a lot more than this; a tool joins the list once I’ve made something real with it.
think & build
Claude
remember
claude-obsidian · Open Notebook
make music
ChatCut · Mureka · ElevenLabs
make video
ChatCut · OpenArt
ChatCut or HyperFrames Studio?
Checked 31 Oct 2026
I use both, for different jobs. ChatCut edits footage and makes media. HyperFrames Studio builds a video out of designed scenes, written as code.
ChatCut
HyperFrames Studio
what it is
A video editor you drive by chat, with a timeline of real clips. Desktop app, web, and inside ChatGPT
HeyGen’s free desktop app. An AI agent writes each scene as code, and the video renders on your own computer
best at
Cutting real footage, songs, AI images and video clips
Motion graphics, explainers, lyric and karaoke videos, text that has to stay sharp
how you review
Play the timeline
Approve a storyboard of stills before anything moves, then comment or draw on the frame
captions
A catalogue of caption templates, some animated. Word-by-word reveal, keyword highlights, two-language captions, Simplified or Traditional Chinese new, not tried yet
91 caption styles and a library of 726 scene ideas
remake a video
Remake a 4–30 second clip with new people, products or a new look, keeping its motion and camera moves new, not tried yet
In the app, but switched off for now
edit by transcript
Yes: change the words and the cut follows
In the app, but switched off for now
screen capture
Records a window or the whole screen into a project and logs every click and scroll, for click highlights later. You or an AI agent drive the app new, not tried yet
Computer use: the agent drives the app and records it (new, on some accounts)
voice, lip sync
Voices, lip sync, avatars and video translation
HeyGen voices, video clips and lip sync
cost
Each generation uses ChatCut credits
Free HeyGen account; the agent bills to its own plan. HeyGen voices, clips and lip sync use HeyGen credits
ChatCut
Songs, AI clips, video built from real footage (talking heads, event recordings), or remaking a short clip.
HyperFrames
Explainers and lyric videos, where the text and layout matter. Two ways to use it.
both
Make the song in ChatCut, then build the video around it in HyperFrames. Several of my music videos started this way.
Both send some data off your computer: ChatCut’s generations run in its cloud, and Studio sends usage data to HeyGen unless you switch it off. For screen recordings, use test data, not real customer screens.
It all started with haze. In October 2026 I built a Singapore haze page and needed somewhere to put it, so I chose GitHub. A haze page can’t live on its own, so it got a home page. Then the home page got tools, videos, songs, an otter, a capybara and a cat.
It’s for SME owners, people learning AI, anyone checking what I can do, and friends and family who are only here for the songs. It’s my own learning in public, made with a lot of help from AI, and separate from my day job.
