Inspiration
One year ago, I was working weekend shifts at a busy restaurant and takeaway. I was 16. I was the bar staff.
Every single shift, two members of staff were permanently pinned to the phones — just taking orders. Not cooking. Not serving. Just answering calls. And they still couldn't keep up. The line was constantly busy. Customers calling in couldn't get through. Customers sitting inside couldn't get served because the staff who should have been on the floor were stuck on the phone.
I'd be behind the bar with a full sheet of drinks to make, and I'd have to leave it, walk out onto the floor, take a table's order myself, go back, make the drinks, and start again. Every night. Controlled chaos.
At 16, standing behind that bar, I thought: this is a solved problem everywhere else. Why is the phone still doing this to restaurants?
That question became Order Flow.
What It Does
Order Flow is an AI phone agent built specifically for restaurants and takeaways. It answers every inbound call, takes complete orders, books tables, sends SMS confirmations, and collects payment — 24/7, including when the restaurant is closed.
The customer calls the restaurant's existing number. Nothing changes for them. Order Flow picks up instantly, handles the full conversation naturally, and delivers the order to the kitchen. No missed calls. No rushed staff. No mistakes.
Powered by Gemini, the voice conversations are natural and accurate — not robotic, not scripted. The AI adapts to each restaurant's exact menu, hours, and rules within 24 hours of signup. No hardware required.
How We Built It
- Gemini handles the core conversational AI — understanding orders, handling edge cases, confirming back to the customer naturally
- Voice telephony layer connects inbound calls to the Gemini pipeline in real time
- SMS confirmation fires automatically after every order or booking
- Payment collected via secure link sent by text before the order is prepared — cutting no-shows and unpaid orders
- Restaurant dashboard gives owners full visibility over every call, order, and booking
Challenges We Ran Into
Real phone conversations are unpredictable. Customers change their mind mid-order, ask about allergens, want to substitute items, or go completely off-script. Building a system that handles this gracefully — without the customer ever feeling like they're fighting a machine — was the hardest engineering problem we faced.
Gemini's reasoning capability was what made it solvable. Rule-based systems couldn't do this. The model needed to understand intent, not just keywords.
The second challenge was trust. Restaurant owners are busy, skeptical, and have zero tolerance for errors when someone's dinner is on the line. Every design decision — from order confirmation readbacks to instant SMS summaries — was built around earning that trust from the first call.
Accomplishments That We're Proud Of
- Built a fully functional AI phone agent as an 17-year-old, self-taught, with no co-founder and no external funding
- Designed a product that customers often don't realise is AI — the conversations are that natural
- Went from idea to working product while studying full time
- Identified and validated a real gap in a market that tech has largely ignored — the 90,000+ independent food businesses in the UK still running their phones manually
What We Learned
AI-native doesn't mean AI-assisted. Order Flow doesn't use AI to help a human do the job faster — it replaces the task entirely. That distinction changes how you build, how you sell, and how you measure success.
The hardest part of building an AI business isn't the model. It's the integration layer — making AI work reliably in a real-world environment where mistakes have real consequences.
I also learned that the best product ideas don't come from research reports. They come from lived experience. I saw this problem at 16 with my own eyes, and that gave me something no amount of market research could: complete clarity on exactly what needed to be built.
What's Next for Order Flow AI
The UK independent restaurant market is the beachhead. Once the model is proven there, the same infrastructure works for any food business — chains, franchises, hotel restaurants, catering companies.
Beyond food, the underlying system — an AI agent that handles inbound calls, captures structured data, and triggers actions — applies to any small business that still runs on phone calls. Salons. Clinics. Estate agents. The market is enormous.
We're currently working with a select group of UK restaurants as design partners — refining the product in real environments, with real orders, before full commercial rollout. The unit economics are proving out. The next step is scale.
Log in or sign up for Devpost to join the conversation.