About Sommo

Sommo started with a model, not a pitch deck.

December 2025: I was deep in fine-tuning experiments, training a model on wine. The first version was promising in a way that made me sit up. The second was top-notch enough that I stopped treating it as a research artefact and started treating it as a product. By the end of the year there was a working demo app, just one feature: scan a label, OCR the text, ask the model what’s in the glass. That was Sommo on day one.

Everything between then and the app you can download today is the work of a community: the friends who tested the demo, the WSET tutors who wrote in with feature requests, the early users who flagged bugs and edge cases and “wait, the cellar should really do this”, the family members who patiently nodded through dinner-table demos. I take every piece of feedback seriously. Wine Character Analysis came from an sommelier who wrote in. Tasting mode came from an active user who shared their experience. Typed-mode WSET grading came from another user who said the choice-mode quizzes weren’t enough. The roadmap isn’t really mine alone.

Underneath the surface the engine kept getting better too. The AI inside Sommo is one I own, not one rented from a frontier provider. I keep pouring my technical knowledge into it, refining it, retraining it, week after week. That’s how Sommo can stay this cheap and still get smarter every release.

Below are six things I believe about wine apps, and the things I actively refuse to do because of them.

i. A wine app shouldn’t cost more than the wine.

Most “premium” wine apps charge $5 to $10 a month and stop short of recommending anything specific. Specialist wine-study apps charge $50 to $200 per WSET level. You could pay several hundred dollars a year to scan, journal, and study using three apps that don’t talk to each other.

Sommo Premium is $2.50 a month if you pay yearly, or $5 a month if you pay monthly. That’s the cost of running the model and the infrastructure, plus a small margin to keep building. Charging more would feel dishonest given the structure.

What I won’t do. No upsell modals. No price-test surge pricing. If you subscribe at one rate, you stay there for as long as you stay subscribed, even if the price goes up for new sign-ups.

ii. The AI should reason like a sommelier, not a chatbot.

Most wine apps that mention “AI” mean a generic chatbot dressed in wine colours. Confident, fluent, frequently wrong. Sommo runs on a model I built and own, trained on wine instead of the open internet. Every recommendation is checked against a wine library I maintain by hand, so the wineries Sommo names are wineries that actually exist.

It’s a growing AI: I keep adding data, fixing mistakes, and refining how it reasons. Because the model is mine, not rented from a frontier provider, every query Sommo makes is a query I’m not paying someone else for. That is the entire reason Premium can stay where it is.

What I won’t do. I don’t ship “AI features” that generate confident tasting notes from a grape name and a vibe. If the model doesn’t have grounded data for a wine, the app says so.

iii. Your journal is yours.

Sommo stores your tasting notes encrypted. They power your wine character analysis, which builds a personality profile from your last ten, fifty, or three hundred entries. That data is never sold, shared, monetised, or used to train third-party models.

If you downgrade from Premium, your journal isn’t deleted, just gated. Resubscribe and the full history comes back. I don’t destroy your data to coerce subscriptions.

What I won’t do. No data brokers. No affiliate “you’d love this” emails. No notifications selling you wine subscriptions because the app noticed you logged three reds last week.

iv. Every screen should feel like a notebook.

Wine deserves quiet. The Sommo interface tries to feel like a leather-bound notebook with very good search, not a sports app, not a slot machine, not a tile-based dashboard. The serif type, the warm palette, the unhurried animations are all chosen because wine drinking is a slow act.

If a feature requires more than three taps to reach, it goes back to the drawing board. Speed is respect.

What I won’t do. No leaderboards, no slot-machine animations, no celebration confetti for ordinary actions. Open Sommo when you have wine in front of you. That’s the right time.

v. The next feature should come from a user.

Sommo’s roadmap is partly mine, but increasingly the people using the app. Premium subscribers vote on what’s next. Wine Character Analysis came from a Premium thread. Tasting Mode came from a sommelier who wrote in. Typed-mode WSET grading came from another user who said the choice-mode quizzes weren’t enough on their own.

There is no product manager between you and the commit log. As a user, you get a seat at the table.

What I won’t do. No “growth features” that exist purely to increase engagement metrics. Every feature has to make at least one user’s wine life genuinely better, or it doesn’t ship.

vi. The person making it should love wine.

Sommo is built by one person, Gökhan, in London. Years of drinking and reading about wine. Two of writing this software in evenings and weekends. There is no VC clock and no board. The app exists because the maker wanted it to exist.

That’s all I’ll say about it. The wine is more interesting than the founder. Always.

What I won’t do. No “founder TikToks.” No selling the company before it’s ready. No raising money just to “scale.” This is a craft business, and intends to stay one.

The governing sentence

A wine app should make the next bottle better than the last. If it does anything else, it’s in the way.

From the founding notes, December 2025.

On the engine under the hood

This is the part most apps wave their hands at, so a quick note on what’s actually inside (in plain language).

Sommo runs on its own AI. I built it, I host it, I keep tuning it. It is not a generic chatbot wrapped in a wine logo, and it is not a frontier model rented per question. It is a wine-specialist AI that lives on my infrastructure.

It’s trained on wine, not the open internet. A curated mix of professional wine writing, structured tasting records, regional and grape data, and food-pairing logic. The persona is fixed: an expert sommelier with decades of selection and pairing experience, never a generic assistant pretending to be one.

It gets better every week. Because the model is mine, I can keep training and refining it without asking permission or paying a per-query toll. New data goes in. Mistakes get fixed. Reasoning gets sharper. That work compounds.

Every recommendation is grounded. Underneath the model sits a hand-maintained library of real wines, wineries, regions, and grape varieties. The app checks the model against the library before naming a bottle, which is why Sommo doesn’t invent châteaux that don’t exist.

It’s cheap because I own it. Most apps that advertise AI features pay a frontier provider per query, and pass that cost (plus a margin) on to you. I don’t. The price you pay is what it costs to run, plus a small margin to keep building.

If you want the technical write-ups, the Sommo model and the wine knowledge graph are both documented on my personal blog. The graph pipeline is also open-sourced on GitHub for anyone who wants to inspect the engineering.

On the founder

I’m Gökhan Arkan, a software engineer based in London. Sommo is the product of two long-running obsessions: wine and AI. Years of drinking and reading about the first, years of building with the second, and at some point in late 2025 the two collided. I started a fine-tuning experiment, training a model on wine. The second pass genuinely surprised me, and by the end of the year there was a working demo: scan a label, OCR the text, ask the model what’s in the glass. Everything since has been built one feature at a time, almost always in response to someone using it and telling me what was missing.

I could have raised money. I could have hired a team. Instead, and this is a deliberate choice, Sommo is going to remain a one-person craft business for the foreseeable future. It is more important that this be good than that it be big.

When I take feature requests from users in the Premium thread, I’m not running them past a product manager. I read them, reply personally, and often commit the change the same week.

If you’ve used a build of this app at any point this year, sent a screenshot of a bug, or argued with me about whether the cellar should sort by drinking window or vintage by default, thank you. The app is more yours than mine at this point.

Where Sommo is, by the numbers

• One person building, running, and supporting the app full-time
• Thousands of active users across iPhone and iPad, organic only
• Five-star App Store rating
• Zero VC raised, no board, no growth team, no clock
• One wine-tuned AI, owned outright, not rented from a frontier provider
• Nine core features, all wired into the same model and the same palate (yours)
• Thousands of WSET flashcards across levels 1 through diploma

Get in touch

Questions, feedback, or just want to chat about wine? I’d love to hear from you.

• Email: hello@sommo.app
• Website: gokhanarkan.com

If anything you’ve read here resonates, start with the free tier. If you want to support this kind of independent work, Premium is $2.50 a month billed yearly. That’s roughly one cheap bottle a year, for an app that probably touches every other bottle you open.