AI Nutrition Assistant

Staying consistent with healthy eating habits is challenging, especially for individuals navigating dietary restrictions or trying to achieve specific healthy goals. Many find the overwhelming number of meal options and nutritional information daunting, leading to confusion and frustration. Traditional meal planning often lacks personalization and accessibility, making it difficult for users to make informed decisions about their diets.

The AI Nutrition Assistant provides a conversational AI that simplifies meal planning and nutrition guidance. By understanding users’ dietary preferences, the assistant offers tailored meal suggestions, easy-to-follow recipes, and instructions.

Project overview

The Nutrition assistance is a RAG application designed to help users develop healthier eating habits, and simplifies the process of meal planning, making nutrition management more accessible and engaging.

The main use cases include:

1. Personalized Meal Recommendations: Suggest meals based on dietary preferences (e.g., vegan, gluten-free)
2. Nutritional Guidance: Providing nutritional information for selected meals (calories)
3. Recipe Retrieval: Filter recipes by meal type (breakfast, lunch, dinner) and cuisine.
4. Interactive Q&A: Answer user questions about nutrition and healthy eating in real time.
5. Cooking Instructions:Offer step-by-step cooking instructions for selected recipes.

Dataset

The dataset used in this project contains 7 columns which are:

• recipe_name,ingredients,nutritional_information,dietary_tags,meal_type,cuisine_type,instructions

The dataset was generated using ChatGPT and contains 180 records.

You can find the data in data/data.csv.

Technologies

• Python 3.11.9
• Docker and Docker Compose for containerization
• Minsearch for full-text search
• Flask as the API interface
• Grafana for monitoring and PostgreSQL as the backend for it
• OpenAI as an LLM

Running the application

Database configuration

Before the application starts for the first time, the database needs to be initialized.

First, run postgres:

docker-compose up postgres

Then run the db_start.py script:

pipenv shell

cd nutrition_assistant

export POSTGRES_HOST=localhost
python db_start.py

Running with Docker-Compose

I recommend to run the app with docker-compose:

docker-compose up

Running with Docker (without compose)

Sometimes you might want to run the application in Docker without Docker Compose, e.g., for debugging purposes.

First we run Docker Compose

Next, build the image:

docker build -t nutrition_assistant .

And run it:

docker run -it --rm \
    --network="nutrition-assistant_default" \
    --env-file=".env" \
    -e OPENAI_API_KEY=${OPENAI_API_KEY} \
    -e DATA_PATH="data/data.csv" \
    -p 5000:5000 \
    nutrition_assistant

URL=http://localhost:5000
QUESTION="Is the Vegetarian Chili recipe suitable for a vegan diet?"
DATA='{
    "question": "'${QUESTION}'"
}'

curl -X POST \
    -H "Content-Type: application/json" \
    -d "${DATA}" \
    ${URL}/question

ID="7cae0f12-ae8b-4a5c-ad77-77e06e00bfef"
URL=http://localhost:5000
FEEDBACK_DATA='{
    "conversation_id": "'${ID}'",
    "feedback": 1
}'

curl -X POST \
    -H "Content-Type: application/json" \
    -d "${FEEDBACK_DATA}" \
    ${URL}/feedback

Interface

Used Flask for serving the app

Ingestion

The ingestion script is in ingest.py.

Used minsearch, as our knowledge base.

Experiments

For experiments, I used Jupyter notebooks. They are in the notebooks folder.

To start Jupyter, run:

We have the following notebooks:

• ai-nutrition-assistance.ipynb: The RAG flow and evaluating the system.
• data-generation.ipynb: Generating the ground truth dataset for retrieval evaluation.

Retrieval evaluation

Used minsearch and below are the metrics:

• Hit rate: 97%
• MRR: 91%

The improved version (with tuned boosting):

• Hit rate: 97%
• MRR: 95%

RAG flow evaluation

We used the LLM-as-a-Judge metric to evaluate the quality of our RAG flow.

For gpt-4o-mini, below is the performance

• 95% RELEVANT
• 5% PARTLY_RELEVANT

Monitoring

WIP

Running it with Docker

docker-compose up