Co-working Fri's First Demo-day

huangpf
Feb 15
6 min read

I almost cancelled Demo Day. Not because something went wrong. Because I was scared my demos weren't good enough. That the whole thing would feel flat. That people would walk away thinking, what was the point of that?

But I caught myself. I run a space where I tell people every single week to build in public, to ship messy, to show up before they're ready. And here I was, about to cancel because my own project wasn't polished.

So I didn't. I went first. I demoed Kai, my half-finished AI community manager in front of everyone. Bugs and all.

A couple of people came back to me this was one of the best Fridays they've had. Because the energy in the room wasn't "look how impressive we are." It was: I'm building this messy thing and I'm going to show you anyway. "Because it allows for people to be themselves and experiment.". Thanks to those who gave me the affirmation, it means a lot.

Anyway, so here's what they built, what broke, and what you can steal from each one.

Demo 1: An AI Community Manager That Calls You on the Phone

Builder: Michael (Me!)

What it is: Kai is a voice AI agent that calls community members, has a real conversation, extracts structured data, and matches them with other members they should meet.

The problem it solves: Over 10,000 people have passed through SQ Collective in the past year. Connecting them manually doesn't scale. So Michael built an AI that does the introductions.

What made it interesting:

The call feels like a conversation, not a survey. Michael used progressive disclosure — the AI doesn't dump all its questions upfront. It lets the conversation build naturally, layer by layer. You reveal more as you get comfortable. Just like talking to a real person.
It remembers you. If you've talked to Kai before, it knows. No repeating yourself. Michael's take: "If you have to repeat yourself, it totally spoils the experience."
It knows when to shut up. Trigger-based call endings mean when you say "I gotta go," the AI actually stops talking. No desperate "But wait, one more thing!" energy.
80+ iterations with a 5-dimension scoring system. Michael built an evaluation/simulation framework across conversation quality, discovery, data accuracy, match quality, and compliance. The big lesson: AI keeps optimizing one dimension at the expense of others. The human has to step in and reprioritize.

Takeaway for builders: You don't need to be technical to design good AI experiences. The hardest part isn't the code, it's the UX decisions. How does the conversation flow? When does the AI stop? What data actually matters?

Demo 2: An AI Interview Prep App Built With Xcode

Builder: Tanmay (Masters, NTU)

What it is: A mobile app that uses your phone's camera and microphone to simulate interview scenarios with AI-generated questions in real time.

What made it interesting:

Built using Apple's native AV signal inputs through Xcode, no stored voice data, just real-time transcription.
Uses PostHog for analytics and Canny for user feedback — two tools every indie builder should know about.
The biggest technical challenge: reducing latency on generating questions from the camera view.

The room's challenge to the builder: Someone in the audience said, "Let's make sure by next Friday, there's a few pieces of feedback on that Canny board." The whole room committed to trying the app and giving real feedback. That's what happens when you demo to builders instead of spectators.

Takeaway for builders: Ship without analytics and you're flying blind. It doesn't have to be fancy, just set up PostHog (it's free) and a feedback board before you launch. Future you will thank present you.

Demo 3: A Tiny Tool That Turns One URL Into a Full Real Estate Listing

Builder: Tony Pelosi (investor and venture builder)

What it is: A boring tiny tool is a small, bespoke workflow built in a few hours. Give it a single property listing URL and it scrapes the data, creates or links to existing buildings, districts, and agents in Airtable, and adds the listing to a calendar-ready database.

Why this demo mattered:

It's not a product. It's not for sale. It's a tool built for one person's exact workflow. And that's the point.
One URL in, fully populated listing out. No manual data entry. No copy-pasting between tabs.
Same approach works for job applications, research, lead tracking — anything where you're manually moving data from a webpage into a spreadsheet.

The philosophy: "I like to show things I only spent a few hours doing, just to show you. If you've got similar workflows, create your own little tiny tools."

Takeaway for builders: Stop looking for the perfect tool. Build a tiny one that does exactly what you need. It doesn't need users. It doesn't need a landing page. It just needs to save you time today.

Demo 4: A Telegram Bot That Turns YouTube Videos Into a Personal Skill Library

Builder: Sophie (first-time builder, Product Manager)

What it is: A Python-powered Telegram bot that takes a YouTube link, transcribes the video, extracts the tools and workflows mentioned, and saves structured notes into Obsidian.

The backstory: Sophie met a community member named Troy a couple of days before the demo. Joy walked through some AI workflows. Sophie got inspired, learned about AI Nation on YouTube, picked up Python, and built the whole thing. In two days.

How it works:

Send a YouTube link to the Telegram bot
Bot transcribes and analyzes the video
Extracts key workflows, tools, and resources mentioned
Saves structured notes into Obsidian with connected nodes

The long game: Sophie isn't just collecting notes. The idea is to build a searchable skill library. "Right now, everything looks like a nail because I only have a hammer. But if I accumulate skills and tools, I can ask AI, "go back to my skill list and tell me which tool is most relevant to solve this problem."

Takeaway for builders: You don't need to be a developer to build your first automation. Sophie went from zero to working bot in two days. Also — stopped building for two months, got re-inspired by a single conversation, and shipped again. That's how it works. You stop. You start again. The community is what pulls you back.

Demo 5: The Water Demo That Started Figma (And What's Coming Next)

Builder: Farhad (community builder at webgpu.com, ex-Mckinsey)

What it is: A tour through the world of WebGL and WebGPU — the technology behind every impressive 3D experience you've ever seen in a browser.

The origin story that hooked the room: Evan Wallace, one of the technical founders of Figma, built a real-time water physics demo using WebGL. He showed it to his co-founder. The co-founder had a moment: "Photoshop can be done in the browser." That demo started Figma. Farhad showed the same demo, now rebuilt with WebGPU — far more powerful.

What Farhad showed:

Real-time physics and caustics running at 60fps in a browser
An educational platform visualizing how LLMs work step by step — interactive, not static
Decentralized GPU clusters in the browser — access someone else's GPU over a network for your own computing
McLaren's website of the year — 60fps 3D rendering, all WebGPU
Digital reconstruction for blast engineers — like Photoshop but for construction sites

What Farhad is building: Interactive video tutorials where you don't just watch, you interact with the content. Inspired by 3Blue1Brown on YouTube, but designed so non-programmers can create educational explainers too.

Takeaway for builders: WebGPU is now default in all major browsers. If you're building anything that touches 3D, AI/ML in the browser, or interactive educational content, this is the stack to watch.

Look at the five demos again:

A voice AI for community matching
An interview prep app
A tiny real estate tool
A YouTube-to-knowledge-base bot
A WebGPU interactive learning platform

Nothing in common, right? Except this: every single one was built by someone who didn't wait until it was ready. The AI had bugs. The app had no analytics. The tool was built in hours. The bot was two days old. The platform doesn't have its animation engine yet.

And every single demo made the room better.

That's what Demo Day is for. Not showing off. Showing up.

Build Something This Week

You don't need a finished product. You don't need a pitch deck. You don't even need to know what you're doing. You just need something you can open on your laptop and say, "I built this. It's not done. But here it is."

Next Friday, the room will be waiting.

Ship first. Then harden.

Missed out last week?

Don’t worry, these conversations happen every Friday.

Usually over laptops. Sometimes over pizza.

You’re welcome to join the next one.