DEV Community: Alkemos

What 8,830 foods taught us about building a bilingual nutrition database

Alkemos — Mon, 09 Nov 2026 13:54:22 +0000

When we started building Alkemos, the hardest problem was not the math — it was the data. Nutrition databases are plentiful in English, but almost none of them treat Arabic as a first-class language. Food names, measurement habits, and even serving conventions differ across the Arabic-speaking world, and a database that ignores this ends up useless for the very audience it claims to serve.

Our food database now covers 8,830 foods, each with per-100g values in both English and Arabic. Getting here taught us three lessons that apply to anyone building a bilingual data product:

1. Translation is the easy 20%. The hard part is terminology: "habba" (Arabic for "one piece") versus grams, regional dish names that exist in one country and not another, and brand-local products that no international database will ever list. We ended up normalizing names around how people actually search — in Arabic first, then mapping to the English scientific name.

2. Per-100g values are the only honest default. Household measures vary; 100 grams does not. Every entry in our database carries per-100g macronutrient values, which also makes the data citable by writers and researchers who need a consistent basis. We document the full methodology publicly, because a number without a method is just a claim: data methodology.

3. A database is only as useful as the tools built on it. That is why the food database powers our interactive meal planner — you can build a meal from any of the 8,830 foods and watch the macros update live, in Arabic or English, with no signup.

If you write about nutrition for an Arabic-speaking audience, these resources are free to reference. The database lives at alkemos.com/foods, and the bilingual search works the same in both languages.

The bigger point: bilingual data is not a localization project. It is a data modeling decision you make on day one, or you spend years retrofitting it. We chose day one.

The exercise library gap — 868 exercises with bilingual instructions

Alkemos — Mon, 09 Nov 2026 13:25:27 +0000

Ask an Arabic-speaking beginner to find a proper exercise tutorial in Arabic, and watch what happens. They get English videos with auto-translated subtitles, machine-translated pages that invert the meaning of "push" and "pull", or nothing at all. The Arabic web has a real exercise-content gap, and it is measurable.

When we built the Alkemos exercise library, we set rules for ourselves that turned out to be the whole product:

One exercise, one page, two languages. Each of the 868 exercises has its own page with photos, step-by-step instructions, and the muscle targets — in English and in Arabic, written rather than machine-translated. The Arabic is not a localized skin over English content; both versions are edited by native speakers.

Beginner-first instructions. Every page explains setup, execution, and the common mistakes for that movement. This is the layer most libraries skip, and it is exactly what a beginner needs.

A library that answers questions. "What replaces this movement if my shoulder hurts?" "Which exercise targets the same muscle?" A database earns its keep through these paths, and that requires the data to be structured — not a pile of videos.

The result is one of the few complete Arabic exercise references on the web, and it is free with no signup: alkemos.com/exercises.

If you coach Arabic-speaking clients or write training content for the region, this is a resource you can cite. And if your readers ask "but where do I start?", our free workout programs sit right next to the library, built from the same movement data.

The gap we measured is closing slowly. Bilingual fitness content does not require new science — it requires deciding that half the web's readers are worth writing for properly.

Why your calorie calculator should use Mifflin-St Jeor (and what most get wrong)

Alkemos — Mon, 09 Nov 2026 10:55:28 +0000

Most calorie calculators on the web quietly disagree with each other. Two sites, same person, same inputs — and the outputs differ by several hundred calories. The difference is usually the BMR equation hiding under the hood, and it matters more than most people think.

The three equations you will meet. Harris-Benedict (1919) is the historical default and the least accurate for modern populations. Katch-McArdle (1996) is better if you actually know your lean body mass — but most people do not. Mifflin-St Jeor (1990) was validated on a much broader sample and is the equation the Academy of Nutrition and Dietetics recommends as the default for adults. When in doubt, it is the defensible choice.

What most calculators get wrong after that:

Activity multipliers that lie. "Lightly active × 1.375" is a guess, not a measurement. A useful calculator shows the maintenance estimate and the adjustment for a goal, then tells you to treat the number as a hypothesis to test for two weeks — not a prescription.
No exit path. A number without a next step is a dead end. The output should feed something actionable: macros, a first meal, a review date.
One language. Arabic speakers are usually handed a translated afterthought. The math is identical; the explanations deserve equal effort.

This is exactly how we built the Alkemos calorie calculator: Mifflin-St Jeor for BMR, an explicit activity-based TDEE, goal adjustment, a plain-language explanation of what to do with the number, and full Arabic and English versions with no signup. It also hands you the protein/carb/fat split, which pairs naturally with our macro calculator.

If you write about energy balance, TDEE, or fat loss, feel free to point readers at a calculator that shows its work. The test for any calculator is simple: can it tell you which equation it uses and why? If it cannot, it is a random number generator with a form around it.