NutriGhar: A Personal Nutrition Companion Built for My Sister and Best Friend
What I Built I built NutriGhar for my sister and best friend, Ayushi Sinha . Ayushi wanted to lose some weight, get fitter, and build a little more shape, and she asked me if I could make something that would help her keep track of her daily calories and protein without making the process complicated. That conversation became NutriGhar. NutriGhar is an offline-first nutrition and…
NutriGhar is a personal nutrition companion app designed specifically for Ayushi Sinha, the reporter's sister and best friend. Ayushi wanted a tool to help her track her daily calorie and protein intake in a simple, non-complicated way as she aimed to lose weight and improve her fitness. The app is built with Flutter and Dart, and follows an offline-first approach, meaning it can function without an internet connection.
One of the main challenges in creating NutriGhar was the way people describe their meals. Indian food descriptions are often informal, mixed with English, Hindi, and Hinglish. Traditional rule-based parsing methods struggle with these varying descriptions, so the reporter focused on integrating an open-source AI layer to improve the meal-understanding process.
The AI component of NutriGhar uses an open-weight model called Gemma 3 4B, which runs locally through Ollama. This allows the app to process meal descriptions without sending sensitive user data to a third-party AI provider, ensuring a stronger privacy model. The AI takes a natural-language meal description as input and extracts structured information, such as food items, quantities, portion sizes, and additional notes.
It then uses NutriGhar's local food database to compare the AI's output with relevant food candidates, grounding the model's output in accurate nutrition values and arithmetic.
The pipeline of the app is as follows: the user inputs a natural meal description, which is then processed by Gemma to extract structured information. The AI then retrieves food candidates from the local food database, grounds the AI's output against these candidates, and finally performs deterministic nutrition calculations. This allows the app to update the user's daily calorie and protein totals and provide insights into their progress towards their fitness goals.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
