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 weight-management app designed around Indian food. It lets a user track meals, calories, protein, weight, goals, activity, and progress while working with foods and portions that actually make sense in an Indian household.
For this challenge, I extended NutriGhar with an open-source AI layer that can understand natural meal descriptions and convert them into structured food and portion information before passing them to the application's nutrition engine.
The idea is simple:
You describe what you ate naturally. NutriGhar does the tracking.
Demo
WhatsApp Video 2026-10-05 at 04.09.24.mp4 - Google Drive
drive.google.com
A typical interaction looks like:
"2 roti, aadha katori rajma, thoda paneer aur ek glass chaas"
Instead of manually searching for each food, the AI interprets the meal, identifies the relevant foods and portions from NutriGhar's food database, and converts them into structured entries.
The app can then update the user's daily calorie and protein totals and help them understand what they still need to eat.
Code
ArmanSinha7
/
NutriGhar
HacktoberFest work week 1 based on a app built for a friend theme helping to make a friend do well on their weight loss and muscle gain journey tracking their daily intake properly
NutriGhar 🥗
An offline-first Indian nutrition and weight-management companion built for my sister and best friend, Ayushi Sinha.
NutriGhar started with a simple request from my sister, Ayushi: she wanted an easier way to track her daily calories and protein while working toward losing weight and getting fitter.
Instead of building another generic calorie counter, I wanted to make something that understands the way we actually talk about food in India — from roti and dal to katori-sized portions and everyday Hinglish descriptions.
What NutriGhar Does
NutriGhar is a Flutter-based Android application for personal nutrition tracking.
🥗 Food & Meal Tracking
- Search a large Indian-focused food database
- Log meals and portions
- Support common Indian food names and aliases
- Track daily calories and protein
- Add custom foods
🎯 Personal Goals
- Set nutrition and weight goals
- Calculate calorie and protein targets
- Track progress toward personal goals
- Monitor weight trends
📊 Progress &
…NutriGhar is built with Flutter and Dart and follows an offline-first approach.
The project includes an Indian food database, food search, meal logging, portion handling, nutrition calculations, goals, progress tracking, and local persistence.
How I Built It
The original NutriGhar application provided the core nutrition infrastructure: Indian food data, food search, portion handling, deterministic nutrition calculations, goals, meal logging, and progress tracking.
The biggest weakness was the way users had to express what they ate.
Traditional rule-based parsing works for clean inputs, but real people don't always describe food in perfectly structured terms.
They say things like:
"2 roti aur thoda paneer"
or
"aadha katori dal aur ek glass chaas"
and often mix English, Hindi, and Hinglish in the same sentence.
So I focused the new AI work on that exact problem.
I integrated Gemma 3 4B, an open-weight model, running locally through Ollama.
The AI layer takes a natural-language meal description and extracts structured information such as:
- food item
- quantity
- portion/unit
- additional notes
The system then uses NutriGhar's food-search layer to retrieve relevant candidates from the local food database and grounds the model's output against those candidates.
This is important because the AI is not responsible for inventing nutrition values or doing the arithmetic.
The model understands the user's language.
The application performs the nutrition calculations.
The pipeline is:
Natural meal description → Gemma → structured extraction → food candidate retrieval → grounded food selection → deterministic nutrition calculation → daily totals
I also kept the AI layer replaceable so that different open models can be evaluated without rewriting the rest of the application.
Why Does Open Innovation Matter?
Nutrition data is personal.
Meals, weight, fitness goals, and eating habits are information that someone may not want continuously sent to a third-party AI provider.
That's why I wanted the AI component to work locally.
With Gemma running through Ollama, meal descriptions can be processed locally rather than being sent to a proprietary AI API. This provides a stronger privacy model and avoids a per-request API cost.
Open models also give me more control over the system.
I can experiment with different models, adapt the application to Indian and Hinglish food terminology, and potentially fine-tune a model for the specific way people describe their meals.
This matters for phrases such as:
"aadha katori"
"thoda paneer"
"mess wali dal"
"2 roti"
A general-purpose nutrition application doesn't necessarily understand the context behind these expressions.
An open model gives me the ability to improve that understanding without making the entire application dependent on a closed API.
The philosophy behind the system is:
The AI understands the meal. The application does the math.
Prize Categories
Best Use of Gemma
I am entering the Best Use of Gemma category because Gemma 3 4B is used as the open-weight AI model at the core of NutriGhar's natural-language meal understanding pipeline.
Earlier Work & Attribution
NutriGhar began as a personal project inspired by my sister and best friend, Ayushi, asking me for a simple way to track her daily calories and protein.
For this challenge, I expanded that foundation with a new AI-driven meal-understanding workflow focused on making natural food logging easier and more practical.
The existing nutrition infrastructure and food data provide the foundation, while the new work focuses on integrating open AI into the part of the application where language understanding creates the most value.
What Ayushi Thought
I built this for someone I actually care about, so the most important feedback is hers.
The goal of NutriGhar was never just to make another calorie tracker.
It was to take a small problem my sister had and turn it into something she could actually use.
Top comments (0)