h4438 / patetrip

Google Vertex AI hackathon

Food Doctor for Travlers

Travelling means going out to new places and it can be very unfortunate if you happen to eat the wrong food.

We decided to build a project that leverages Google Palm 2 and Text-to-Image model to provide users with food knowledge to avoid allergies.

A safe journey is the best journey.

How to run

You'll need a service account JSON file provided by Google Cloud. And do the following cmd.

# At the parent directory
mkdir .google
cp <your_service_account>.json ./google/service_account.json

You'll aslo need a env inside .google directory. The env file contains:

LANGCHAIN=Langchain_plus_api_key

Install python packages.

pip install -r requirements.txt

How we help users

Most people only know certain food that they are allergic to, but don't relize that the allergens are the real cause. So we will based on this information to ask Palm 2 to determine what allergens causing the allergies.

You can look at how this process will be performed from the picture.

Project overall process

The overall process of this project, from the start to the end.

References

• 
• 

patetrip's Issues

Upgrade qa generation chain

Upgrade generated questions

Model should ask questions to clarify patient eating habits:

• what else do they eat with
• what else do they drink
• cross contamination in everyday activities

upgrade food_search's extraction chain

Problem

The input paragraph is too large for the model to process effectively.
We need to split up the paragraph into chunks, then process them one by one.