Meal Nutrition Helper

Description

A streamlit chatbot that allows you to enter a meal name or short description and then returns the basic nutritional information for that meal such as calories, fat, carbs, and protein.

How It Works

There are four stages to the chatbot:

1. On startup we parse a small list of foundation foods from the https://fdc.nal.usda.gov/download-datasets.html website. We extract the food name, calories, fat, carbs, and protein from the dataset for each entry.
2. We generate embeddings for each description of a foundation food. These are stored in chroma-db which is a small in-memory embedding database for quick retrieval.
3. When a user enters a meal description we use an LLM few-shot prompt to expand the description into a list of possible prepared foods and dishes. Transformer based query expansion is a powerful technique for AI based information retrieval.
4. We then take the expanded search terms and chain them into a new prompt template - also using a few-shot learning method - that tries to reason about the common and most prominent ingredients across the set of meals.
5. We do a K-nearest-neighors vector embedding search against chroma-db to find the closest matching foundation foods by the semantic similarity of their descriptions to the ingredients and return the nutritional information for the top k results.
6. Lastly we aggregate the nutritional information across the top 5 results and return the total calories, fat, carbs, and protein for the meal.

Environment Setup

• Make sure [poetry] is installed(https://python-poetry.org/) - Virtualenv setup
• Make sure python 3.11.4 is installed
• Install python dependendcies:

$ poetry install --no-root

• Spawn a virtualenv

$ poetry shell

• Add a secrets.toml file

$ touch .streamlit/secrets.toml

• Add your personal openai api key to the secrets.toml file you just created

openai_key = "REPLACE THIS WITH THE CONTENTS OF YOUR API KEY"

• Set your PYTHONPATH variable to the repo root directory

$ export PYTHONPATH=$(pwd)

• Run the streamlit app

$ streamlit run src/main.py

• Open your browser to http://localhost:8501 and use the chatbot