ARC Ladies AI Demo Day: Session 1 Recap

Wednesday, June 10, 8:00 PM - 9:30 PM GMT+8

View Luma event

Five ARC ladies shared practical AI projects, from personal tools to portfolio, website, content, and agent workflows.

JoanChristineJeanNatJac

Hosted by Mell

Joan 🇲🇾 — "Ruru Play & Learn," an ABC learning game for her toddler

~00:01:21 – 00:11:27
“Oh, mommy! She can say Ruru!”
  • Key takeaway: With no coding background, she built a working ABC learning game for her ~3-year-old daughter in under an hour, just by chatting with AI.
  • Three modes: Learn ABC, Match (find the small letter), and a Quiz ("A for Apple").
  • Designed around a toddler: bigger letters for small fingers, simpler colours (changed white→yellow for visibility), with plans to add harder "levels" as she grows.
  • Runs on her phone; her daughter plays ~15 min on the way to preschool.
  • Her next goal: an AI agent to book her yoga/reformer classes.

Christine 🇦🇺 — Rebuilding her design portfolio with Claude Code + Figma

~00:11:27 – 00:24:20
“There's a difference between a brainstorming task where you let the AI think with you... compared to an execution task where you just tell it to do this.”
  • Key takeaway: A former Canva engineer (now pivoting to graphic design) rebuilt her portfolio in ~2 hours / 94 prompts by exporting rough Figma screens and letting Claude Code build them.
  • Her old portfolio was hand-coded, took ~3 days, and looked mediocre; she wanted cleaner, mobile-friendly, and easy to add future projects to.
  • Claude nailed the layout in roughly one go; she only adjusted minor spacing/fonts, usually one prompt per change.
  • Tips: start with a clear goal; distinguish brainstorming vs. execution tasks; correct UI/UX terminology gets better results; ask the AI to help write your prompts.
  • Next up: a "digital visual archive" with richer animation and interaction.

Jean 🇹🇭 — Redesigning her company website with Claude Code in the terminal

~00:24:20 – 00:37:41
“I felt like Claude is like an experienced teammate who kind of like expanded on my work already”
  • Key takeaway: A self-described non-developer replaced her 5–6-year-old Squarespace site by giving Claude Code (in the terminal) just her existing URL — it analysed the site, rewrote the copy, and rebuilt it in ~20–30 minutes.
  • Motivation: the old site felt dated and cost ~$290 SGD/year; she wanted more control without a web-design background.
  • Claude added things on its own — extra sections (e.g., a Q&A) and a working contact form via Formspree (she never specified the API).
  • Observations: Claude has a recognisable "house style," asks clarifying questions instead of over-assuming, and acts "like an experienced teammate." She also used co-work to auto-generate a branded PowerPoint.
  • Next: reskin and ship via a host (Vercel/Netlify).

Nat 🇸🇬 — "Product Marketing Voice OS," an AI brand-voice tool

~00:37:41 – 01:02:51
“...like this ended up getting me to the final round.”
  • Key takeaway: Built a marketing “voice OS” that writes and audits on-brand content — and that project helped her land a job at an AI company
  • It scrapes brand references into a "tone library," generates platform-specific posts in the brand's voice, and audits drafts — scoring them, flagging "AI slop" words ("supercharge," "game-changer," "revolutionary"), then rewriting.
  • It was built across three iterations using Zo Computer (a multi-model platform), Claude Code, and Chatgpt.
  • Workflow opinions: ChatGPT for visual/creative, Claude for analysis/criticism/audits; use JSON prompts for images; generate prompts in the main app, then paste into Claude Code to save credits/context.

Jac 🇸🇬 ("Mama Jac") — A Telegram reminder bot + AI agent, built step-by-step

~01:02:51 – 01:31:16
“Treat Claude as your personal IT coach, which can never get frustrated with me”
  • Key takeaway: Her message was that anyone can start — she deliberately built simple things step-by-step, using Claude as a patient "personal IT coach" that never gets frustrated.
  • Project 1: a Telegram bot for small to-do reminders — automation built first on Make.com, then rebuilt on n8n, running on a cheap VPS (she explained VPS vs. buying a Mac Mini, and the data-privacy angle. chose a VPS over a Mac mini because it was far cheaper, just a couple of dollars a month versus a few hundred upfront (Claude had recommended it for exactly that reason)
  • Project 2: an AI agent (via OpenRouter) integrated with Google Calendar to set recurring birthday reminders (1 month / 1 week / day-of) — all by prompting in Telegram; it has since replaced her n8n automation.
  • Practical lesson: progress gradually, start simple, be patient troubleshooting, and force the AI to explain one step at a time.

Glossary

Vibe Coding

A collaborative approach to software development where a user uses natural language and AI prompts to build applications without writing the code manually.

AI Slop

Generic, overused, or hype-driven language commonly produced by AI models that lacks human nuance (e.g., "game changer").

MCP (Model Context Protocol)

A connection standard (such as the Figma MCP) that allows AI agents to integrate with and pull data directly from third-party platforms.

N8N

A node-based automation tool used to connect different apps and services (like Telegram and Google Sheets) into a workflow.

VPS (Virtual Private Server)

A virtualized server sold as a service by an Internet hosting provider, used for running private applications or AI agents.

JSON

A data format used to provide highly specific, structured instructions to an AI, particularly useful for maintaining accuracy in image generation.

OpenRouter

A service that provides access to a vast library of different Large Language Models (LLMs) through a single integration point.

design.md

A plain-text markdown file that describes a brand's visual language in a format AI coding agents can actually use. https://getdesign.md/what-is-design-md

Additional Resources

Joan's prompt library, shared during the session. Timestamp: ~58:23-1:00:00.

Xiaohongshu’s trend prompt

  • Upload photo of yourself on chatgpt and paste this

以上傳照片中的人物為原型，復古膠片攝影，明顯的膠片顆粒感，真實光影，復古 調，富士濾鏡。舊報刊亭一側擺放枯木，旁邊盛開橘粉花叢，復古膠片顆粒紋理交織，虛焦背景拖影殘跡突出，重影效果營造強烈非現實主義氛圍。一位妖冶美少女抱著花側身，冷白皮細膩通透，精緻面容帶動感模糊，人物輪廓隨平移鏡頭微微搖晃，暖色調低飽和濾鏡，自帶柔光的長髮逆光下閃閃發光，凌亂發絲快速飄動拂過下頜，展現喜歡與危險的吸引力，可愛動作讓畫面靈動，運動模糊與膠片顆粒融合，兼具網紅網感與故事感氛圍，平衡妖冶與甜美。面部感覺調整為符合整張圖片。

 

Color analysis’s trend prompt

  • Use chatgpt, upload your photo and provide prompt

Create a personal color analysis graphic using this portrait. Show side-by-side clothing color comparisons to highlight which colors suit the subject best. Make it visual-first, with short labels only and no paragraphs.

Include make up (eye, cheek, lips) and hair colour that suits me vs I should avoid as well

 

Want to share something you've built with AI?

We want to see it! If you or anyone you know has built an app, agent, or workflow with AI and would love to share, drop Mell (@gmmell) a message and we'll find you a spot!