It started with a dish nobody had modelled.

The company

Love Salone — salone is what Sierra Leoneans call Sierra Leone — was founded to eliminate the friction in deciding where to eat, and to turn what it learns along the way into something researchers can use. Those are two halves of one idea. A system that understands a palate well enough to recommend dinner understands it well enough to tell a public-health researcher something true about how a population eats.

The flagship product is TasteTwins, an iPhone app that learns what you like at the level of the dish rather than the cuisine, then finds people whose palate overlaps with yours and helps you make a plan with them.

Why Sierra Leone matters to a taste model

Almost every food recommendation system in use was trained on Western restaurant data. That sounds neutral. It isn’t.

When you build a taste model from American and European menus, the dimensions you end up with describe American and European food well and everything else approximately. Fat becomes one axis, so red palm oil — which carries a distinct, slightly bitter, deeply savoury character central to West African cooking — sits in the same coordinate as olive oil. There is no viscosity dimension, so the difference between a thin broth and cassava leaf stew ground to a heavy paste simply does not exist in the model. Dried fish, used as a background seasoning rather than a listed protein, disappears entirely — which is a flavour problem and, for someone with a fish allergy, a safety one.

The founder’s family is Sierra Leonean. That is not decoration on a company story; it is where the technical insight came from. Trying to describe the food he grew up eating using the vocabulary these systems provide made it obvious how much the vocabulary was leaving out.

So we started there, deliberately, at the hardest end. We have built a seed dataset of Sierra Leonean dishes with their Krio names, their ingredients and their preparation methods, and used it as a stress test against our own taste model — cataloguing every place the model fails to represent something real. Each gap it exposes is a dimension worth having. A model that can tell plasas from a curry can tell far more ordinary things apart too.

The founder

Corey Arnold is the founder and chief architect. He is a DevOps and systems engineer who has spent his career on high-assurance defence platforms inside Sensitive Compartmented Information Facilities, holds an active CompTIA Security+ certification, and builds cloud pipelines and secure integrations for a living.

That background shows up in the product in a specific way: the assumption that data about a person should be protected by architecture rather than by a promise. It is also why the privacy policy says what our security actually is and declines to call it something stronger.

Where things actually stand

We would rather you have an accurate picture than an impressive one.

• The iPhone app is real and works. The screenshots on this site are genuine captures from the current build running on a simulator, not renders or mockups.
• It is in private testing. A small invited group is using it. It is not on the App Store, and there is no public TestFlight link — we are opening access deliberately and in small batches, in step with what the infrastructure can carry.
• The demo on this site is a browser simulation. Its people and venues are invented and its scores are simplified illustrations, not the production engine. It is labelled that way on every screen.
• The catalogue is modest and real. Actual figures are on the technology page. They are not in the millions, and we are not going to say they are.
• The research programme is early. The privacy safeguards a research dataset would require are a design requirement, not a finished implementation, and no data has been shared with anyone.

What we are looking for

• Early testers — particularly people who eat widely, and people whose food is badly served by existing apps. Ask for access here.
• Restaurants, especially West African and other underrepresented kitchens, willing to have their menus described properly.
• Researchers in nutrition, public health and food science who want to talk about what this kind of data could and could not support.
• Partners and press — see the press page, or just write to us.

---

Everything here goes to one place: hello@tastetwins.app. It is read by a person.