.png)
From Idea to App Store with Peter Hanschke
Not everyone knows this, but OPMMA began as a LinkedIn group, and Peter Hanschke is the one who started it. So, it was good to have him back in front of the community this month, talking about how he built myFabWines, an iPhone app that suggests wines for whatever you're cooking or eating in a restaurant. He did it on his own, using AI where he'd normally have needed a developer and a marketing team.
Peter has had a long product career in Ottawa, including JetForm, Corel, Kinaxis and TechInsights, and he started out as a software developer. He's also a trained cook who works in a commercial kitchen and volunteers at the Ottawa Mission. And he likes his wine!
The idea
They say necessity is the mother of invention, and in Peter’s case the necessity was a Christmas turkey. He had bought a good bottle for a family dinner and it tasted bitter with the meal, so they opened something else. After dinner he tried the first bottle on its own and it was excellent.
Most of us can manage steak and a red, but he couldn't find an app that would tell him what to open with chicken cacciatore. The wine apps he found were mostly about the wine, with pairing as an afterthought or tucked behind a subscription.
Checking before building
What I liked about Peter's approach is that he didn't start with code. He worked through the problem first, the way many of us learned on the Pragmatic course, and used ChatGPT as his research assistant.
His first test, in early 2025, was whether the idea was even possible. He asked ChatGPT what to pair with chicken cacciatore and got a decent answer straight away, if a very long one. It took a few rounds of rewording to get it short enough to be useful.
He then gave it a role before each question, "wine market researcher" for the size of the market and "wine iPhone app researcher" for the competition. The second one came back with a table of apps, their user numbers and each app's main feature, which showed him most of them were built around the wine rather than the meal.
Someone asked whether a lack of competition might mean nobody wants the thing. It's a fair question, and Peter's answer was to talk to people. He described the idea to about 15 friends and said if they'd told him it was a bad idea, he'd have built it for himself and left it there. They were keen, and came back with more feature ideas than he could use. As he put it, "At some point, ChatGPT is not going to be drinking the wine or eating the meal."
He also had two restaurant stories that will be familiar to anyone who has stared at a 15-page wine list. A sommelier in Picton got it spot on for him and his wife. A server at an Ottawa restaurant recommended a new arrival that she said went with everything. It didn't, and ChatGPT agreed with him when he checked at home.
His MVP was small. Type in a meal and its sides, and get back a short list of wines with a reason for each, in under 10 seconds. Anything slower gets awkward when you're sitting at a restaurant table with friends.
Building it
When Peter started, the AI handed him code in pieces that he pasted into Xcode, and when it didn't compile he was back on Stack Overflow. He said his developer background got him through. Adding a feature this year was much easier, and he thinks you need far less coding knowledge now than you did when he began.
He used ChatGPT first as a UX/UI designer and then as a developer to write the code. The part I found most interesting was the prompt that sits inside the app. It sets the AI up as a sommelier who isn't a snob, keeps the prices reasonable, and tells it what format to send the answer back in so the app can display it cleanly. Someone in the audience asked whether it's told not to make things up, and Peter said it is.
There's no wine database behind it. The model understands what's in a dish like poutine and what different wines taste like, and matches the two. Ask twice and you'll get slightly different answers, which bothered Peter the computer scientist until it was pointed out to him that two sommeliers wouldn't agree either.
His advice was to build in small pieces. The one time he asked for everything at once, what came back wasn't close to what he wanted.
The demo ran without a hitch, apart from someone in the room suggesting pine needles. That led to a good point from the floor. With no database, the app will always suggest something. Peter has checked a long list of meals against what sommeliers recommend and says the results are generally sound, but I suspect trust is something he'll keep working on. One suggestion from the audience was to let users rate pairings over time.
Getting it out there
Before launch, Peter used Apple's TestFlight to put the app in the hands of 15 or 20 people, and his son, who is a designer, helped with the look. For marketing he moved to Claude, because Projects let him keep all his research in one place and Skills let him reuse his Pairing of the Week posts across platforms. It suggested Instagram and Facebook, since food and wine photograph well. He read and adjusted everything before it went out.
My favourite story of the night was from a stopover in Toronto on the way back from the Caribbean. A woman at the airport bar downloaded the app, liked it, and told him what he needed were micro press releases. Peter had never heard of them. When he got home, Claude's view was that they would help later, once the app is better known.
He was candid that two paid Meta campaigns didn't work and that social media marketing isn't his strength. That's where he leaned on AI the most.
What he took away
Peter's closing lessons were practical. You own everything now, including the responsibility to your users. The AI is your team, so when you need a code change or new copy, you ask for it. Choose the tool that suits the job, and if your employer mandates one, use it as long as it does what you need. And treat it like a colleague, push back on it, and don't take its praise too seriously.
He told a story about a consulting client years ago who funded development a little at a time through friends and family and ran out of time before the product got to market. Peter thinks it would have been built much faster today.
His last slide said, "AI is not a replacement for the PM or PMM. The judgment, the decisions, and the work are still ours to do." Whether one person can grow this into a business is still an open question, he said. Next up is a feature that lets you photograph a restaurant's wine list and choose from what's on it, which could open doors with restaurants and wineries.
At the end of the evening, I ran a quick poll and asked who felt inspired to build an app of their own. A lot of hands went up, mine included. If you’re one of them, ProductCamp 2027 would be a great place to share your journey.
Thank you
The questions didn't stop all night, and as our self-appointed timekeeper, Michael struggled to keep up. Thanks, Peter, for coming back to the community you started and for being so open about what didn't work as well as what did.
Thank you also to Ed Goffin and the team at Pleora Technologies for hosting us at their Kanata office. Having a home for these evenings makes a big difference for a volunteer-run community like ours, and we appreciate it.
