Forky AI journal.
Engineering deep-dives from the team building Forky AI from Lausanne. No diet advice, no SEO filler — just how the pipelines actually work, what we tried, and what failed.
AI calorie counter accuracy: 30 plates, three apps, USDA ground truth
We photographed 30 standard plated meals, scored each one with Forky AI, Cal AI, and Snap Calorie, and compared the results against USDA-derived ground-truth values. Per-component vision wins on plates with toppings; single-photo estimators tie on simple foods. Raw spreadsheet published.
Fridge-to-recipe prompt engineering: how Forky's 3-pass vision works
The actual prompt structure Forky AI uses to go from a fridge photo to a structured ingredient list. Three passes, why each exists, the failure modes we hit, and the JSON schema we hand back to the app. Builder-focused, with code.
How AI calorie counting actually works (and where it breaks)
A look under the hood of photo-based macro estimation — why whole-plate vision drifts ±25%, why per-component decomposition lifts that to ±10–15%, and the failure modes that no AI calorie counter has solved yet.
Importing a recipe from any URL, photo, or PDF — the pipeline
Most recipe-import features hand-wave 'AI' and ship a Spoonacular wrapper. Forky AI's pipeline does four things differently: scrapes structured data first, falls back to vision OCR, generates a hero photo, and vision-verifies that photo before saving.
Tracking shelf life with AI — why the 'expires in 3 days' badge isn't lying anymore
Most fridge-tracker apps store a static integer for expiry. Open the app a week later, the badge still says 3 days. Forky AI's expires_at refactor — and why timezone-aware countdowns matter — explained.