A homegrown phone receptionist now answers thousands of restaurant calls a night, from Bondi to Brunswick. Its creators say getting it to sound human meant scrapping their first version and starting from scratch.
If you rang your local for a Friday night parma in the past year, there is a decent chance the voice that took your order was not a person.
You probably did not notice. That is the whole idea.
Australian restaurants miss a huge share of their phone calls. The line is engaged, the kitchen is flat out, and nobody can break away to grab the handset in the middle of a dinner rush. Every missed call is a missed order, and for a venue running on thin margins, those add up fast.
Dineline AI, an Australian-built artificial intelligence receptionist, was created to make sure none of those calls go unanswered. More than two years and over 200 restaurant installations later, its founders say the road from clever prototype to a system restaurants actually rely on taught them more than they bargained for.
It started as a demo. Then it met a real kitchen
The first version came together in weeks, built in the fast, improvised style now nicknamed "vibe coding": describe what you want to an AI tool, wire up a few services, and watch something that almost works appear overnight.
It demoed beautifully. Owners leaned in. But a demo only has to work once, with someone watching. A receptionist has to work every time, with nobody watching.
"The first version was a salesperson, not a product," the team says. "It proved restaurants wanted this. What it didn't prove was that we could handle every deal, combo and pack, every size and price variation, the meal deal where the drink is only free on weekdays, and run all of it on a wet Saturday night across a hundred venues at once."
That gap forced a two-year rebuild into what engineers call a "native" platform, with developers writing real code for the messy reality of hospitality.
Menus are chaos, and customers don't talk like the demo
The biggest shock was the menu. Feeding one to an AI as a slab of text works fine until the orders get real: half-and-half pizzas, "no onion, extra cheese, gluten-free", daily specials scrawled on a board, allergy flags that are legally serious, and the regular who orders "the usual" and expects to be remembered.
Handling that reliably took a purpose-built ordering engine, not a clever guess from a chatbot.
Then there is the way Australians actually speak down the phone. The demo was tested in quiet rooms by clear voices. The real dinner-rush caller has a TV blaring, kids in the background and an accent from any corner of the country or the world. Off-the-shelf speech tools choked on "parma", "schnitty", "snags" and half the suburb names in the country.
And on a call, timing is everything. People forgive a chatbot that pauses for two seconds on a screen. On the phone, two seconds of silence sounds broken, and the caller hangs up. Closing that gap meant taking control of the audio itself, so the system starts thinking before the customer has even finished talking.
Selling like a good waiter
Owners wanted more than a phone answered. They wanted the gentle upsell a good staffer delivers: garlic bread with that, make it a large, have you tried the new dessert.
That turned out to be one of the hardest things to get right. Push too hard and it nags. Suggest the wrong thing and it costs the venue the customer.
"Sounding warm is easy," the team says. "Knowing the one moment to suggest the cheesecake, and then letting it go, is the engineering."
Why they tore it down and rebuilt
The team is blunt about why the rewrite happened.
"Our first system was vibe and Vapi coded, which let us move fast and prove the concept. But the moment we put it into a real restaurant, the cracks showed," they say. "Real venues are far more complex: noisy calls, impatient customers, item variations, packs, deals, substitutions, upsell rules, printer workflows, fallback handling and peak-hour pressure. A platform built on someone else's tools limited how much we could control."
"If Dineline was going to be a serious operating platform and not just a clever AI demo, it had to be built natively from the ground up. That gives us control over reliability, speed, security and round-the-clock support, and it lets us protect the price."
This was not, the founders stress, a Vapi-coded freelance project stood up in a weekend. It is a ground-up AI application, with more than $1 million invested so far and over two years spent building and installing phone-answering systems in real venues. That experience, they say, is the difference between a demo and something a kitchen can lean on at 7pm on a Friday.
That price is central to the pitch: a flat rate of about $3.50 an hour, no commission on orders, more than 100 languages, every call answered, several at once. Those numbers, the founders say, are impossible to deliver while quietly paying for someone else's services in the background. Owning the technology was the only way the maths worked.
The part customers haven't clocked yet
For all the engineering, Dineline AI's CMO, Kirstie Wood says the real hurdle was never technical. It was comfort.
"People want to hear a familiar voice when they ring their local, and we knew from the start how much that mattered," Miss Wood said. "So we built Dineline to actively mimic the voice of the restaurant owner. Their regular customers get the same experience they have always had, while the system quietly optimises every order in the background."
"And comfort with AI is only growing. We saw the same curve with Uber, with self-service checkouts, with banking on our phones. This is no different, and the upside is bigger than people expect," she said.
She argues the technology actually improves the experience inside the room.
"When a server isn't getting pulled off the floor to answer the phone, they're fully present with the table in front of them. You get more of your waiter, not less."
"Ordering through AI means a restaurant never misses an order just because nobody could reach the phone during the rush, and the customers in the dining room get the full, uninterrupted attention of their server."
"This is the future of restaurant reception. In five years, every restaurant will have one." — Kirstie Wood, Chief Marketing Officer, Dineline AI
Two years on, that future is already answering the phone when you call for your parma. And you would never know.
If you are a restaurant owner, hospitality operator or takeaway venue looking for an AI receptionist, visit dinelineai.com to get your free trial of your customer receptionist.