This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend
What I Built
Meet Sugar
Sugar is my friend. She does nails in GRA, Port Harcourt: gel, acrylic, Gel-X, BIAB, French tips, chrome, little hand-painted flowers. She's good at it, so she's busy.
I've watched the same thing happen at her station many times. She has a client's hand in one hand and a brush in the other. Her phone keeps lighting up: "Hi dear, are you free today? How much is acrylic?" Someone walks in and asks how long the wait is. She can't stop to reply, she can't write down who came first, and she can't ask the next person if they want coffin or almond, nude or wine, or whether they're allergic to anything.
So people wait, some leave, and the order of who's next lives in her head.
She doesn't need salon software. She needs a front desk: someone to greet people, write down exactly what they want, tell them the price, take payment and give her a clean list of who's next. So I built one.
What it does
Two screens.
The customer site. Customers open it on their phone, from a link Sugar shares or a QR code she can print for her station. The front desk greets them and asks for their name and number, then the service, style, colours, length, shape, allergies and a time. They can tap the options or type the way they talk, in English or Pidgin. They see the full order and the total in naira, pay, and get a ticket that updates on its own: "You're #2 · 1 person ahead of you."
Sugar's queue. A PIN-protected page on her phone or tablet. It shows who's in the chair, who's next, and each person's order: services, style, colour swatches, length, shape, any allergy note (highlighted), the price and whether they've paid. She taps Start when someone sits down and Done when they leave. New bookings chime and slide in by themselves.
The front desk is Qwen3-8B, an open-weight model I fine-tuned with LoRA on Tinker. It talks like Sugar's front desk, knows her menu and prices, and ends every booking with a structured order the app can check.
Demo link
https://sugar-nails-nine.vercel.app
Demo video
https://youtu.be/ZyzuW1q0ZAE?si=SgPEC1N9pV0yKqDI
Code
https://github.com/comzzy/sugar-frontdesk
How I Built It
Sugar doesn't need a chatbot that sounds smart. She needs one that gets the order right: the correct service, the correct price, and the correct time, in the way her customers actually talk, which is often Pidgin.
A general model couldn't do that well. When I tested base Qwen3-8B on her menu, it made up prices, forgot to ask about allergies, and sounded nothing like her. So I trained it on Tinker.
How I used Tinker:
- I wrote about 1,750 booking chats in Sugar's voice, in English and Pidgin. Each one ends with a clean order card: name, phone, service, style, colours, length, shape, allergies, time and naira total.
- I fine-tuned Qwen3-8B with LoRA on Tinker. I didn't have to set up any GPUs. The training took under 5 minutes and cost about $1.65.
- I tested the base model and my trained model on 40 chats neither had seen before.
The results:
- Orders that came out fully correct went from 22.5% to 95%.
- Prices that were right went from 35% to 92.5%.
- Order cards the app could actually read went from 72.5% to 100%.
The live chat runs on that trained checkpoint through Tinker. Each reply costs about ₦0.31, and a full booking comes to roughly ₦2.80.
The rest of the app: The site is a FastAPI app on Vercel. When a customer finishes booking, the order card lands in Sugar's queue, a PIN-protected page where she taps Start when she begins a set and Done when she finishes. The payment step is simulated for now.
Why Does Open Innovation Matter?
Sugar is one person with one phone, working out of GRA, Port Harcourt. She's never going to pay for an enterprise AI subscription, and no big company is building a booking assistant that speaks Pidgin and knows her prices.
Open models change that. Because Qwen3-8B is open-weight, I could take it and shape it to fit her exact menu, her customers and the way she talks. Tinker made the training part easy: I didn't need a GPU or a big budget, just good data and a few minutes. The whole fine-tune cost less than a plate of food.
That's what open innovation means to me. The tools aren't only for big tech anymore. Anybody who knows a small business well enough can build something that actually fits it, and keep full control of it. The code is open too, so another nail tech, barber or caterer can fork it, swap in their own menu, and have their own front desk.
My Agent Session
https://dev.to/agent_sessions/building-sugar-nails-nocuim
Prize Categories
Best Use of Tinker.
Top comments (0)