maken.fit

I built maken.fit to make nutrition tracking feel more like a conversation and less like filling out a spreadsheet. People can describe what they ate, upload a meal photo or nutrition label, and review an editable meal draft with the foods, portions, and nutrition broken down.

The processing pipeline uses OpenRouter and TypeSafe’s Jev to turn a meal description into decisions the application can work with. Jev classifies supported entries and selects matching food references and serving measures. Packaged USDA data and product references supply nutrient values, with OpenRouter web search providing evidence when branded foods need a lookup. Python handles portion scaling and totals. When a match is unclear or a correction needs more context, the pipeline falls back to a language model instead of forcing a result.

I built the review flow to keep people in control of what gets logged. They can change quantities, correct a food, or give a follow-up instruction before saving the meal. The draft retains source evidence, assumptions, and uncertainty, and only an explicit save adds it to their journal. Photo and label inputs feed into the same processing and review flow, so the experience stays consistent regardless of how someone starts an entry.