A second app is easy to imagine and hard to justify. These are the filters we use before an idea earns a slot in the DualGem portfolio — and the reasons we say no to most of them.

ProShot AI is live. App 02 is in development. The third slot is still open — and that’s deliberate. A small studio doesn’t get to build everything it’s curious about. We get to build the things that survive a short list of questions.
This note is that list, written down. If you’re wondering why DualGem doesn’t have ten apps, or why the next one doesn’t have a name yet, this is the answer.
The short version
1. Start from a repeated problem
We don’t start from a technology. “What can we do with vision models?” produces demos, not products. We start from a task people already do — often badly, with a workaround they’re embarrassed by.
The signal is a folder of screenshots and a group chat. A spreadsheet doing a database’s job. A photographer booked three weeks out for a photo that takes three minutes. If the workaround is still alive after the novelty fades, there’s a real problem underneath it.
If nobody repeats the task, the app is a party trick. Clever once, gone by Friday. We pass.
2. Quality must be judgeable
A DualGem app’s output has to be obviously good or obviously wrong. You show someone a result and they know — without being told — whether it worked.
Vague “AI magic” fails this test. If quality depends on the user having taste, or on us explaining why the output is actually great, the product isn’t ready. The boring parts of the flow — onboarding, waiting, paying, keeping the result — matter as much as the model, because they’re where quality gets felt.
If you need a paragraph to explain why the result is good, it isn’t good enough to ship.
3. The business model must be honest
DualGem is bootstrapped. We charge for the product, and that’s the business. So the pricing has to be something we’d be comfortable saying out loud, in plain numbers, on the product page.
That rules out a lot of familiar tricks: fake scarcity, countdown timers, free trials that quietly become subscriptions, “credits” that expire in 48 hours to force a purchase. If the model needs a dark pattern to work, the model is the problem — not the marketing.
Privacy is part of the same test. If we can’t describe what we collect, how long we keep it, and when we delete it in one honest paragraph, we’re not building it. ProShot deletes source selfies within 24 hours and generated headshots after 14 days. That isn’t a policy page talking. It’s the product.
How ProShot passed
ProShot AI started as a frustration: a good professional headshot cost hundreds of dollars or a week of scheduling, and the photos people already had on their phones were usually enough. Repeated task: yes — profiles, CVs and company pages need headshots constantly. Judgeable quality: yes — you can see in a second whether the result looks like you. Honest model: yes — free credits to try, packs that never expire, subscriptions that say what they are.
App 02 is going through the same filter right now. It has a slot, not a name. It gets a name when it’s good enough to show — and not before.
How we say no
Most ideas die in the research slot. They’re interesting, or timely, or popular with a competitor — and they fail one of the three tests. That’s the firewall that keeps the portfolio small on purpose. A focused studio’s job isn’t to ship more. It’s to ship the next thing only when it earns the name.
If you have a problem you’ve been working around for years, tell us about it. The best candidates arrive as complaints, not feature requests.