-
-
Streamlit
Inspiration
New Year Resolutions. We all make them, but drop them because their too much work. Healthy eating is probably on most peoples list, what better way to promote it, then to make it easier to determine the nutritional value of recipes your making.
What it does
A Streamlit app powered by Snowflake and Mistral, designed to simplify meal planning and provide nutritional information.
How we built it
- Cortex Search: Utilizes a RAG engine with cookbook text data for semantic search and contextualized responses.
- Mistral Large 2: Leverages Snowflake Cortex for generation.
- Streamlit: Provides a simple and intuitive frontend for displaying responses.
Challenges we ran into
The biggest challenge was finding the right data to use for RAG and coming up with an app idea that uses that data to improve on normal generation without it becoming a generic Q&A app. Another challenge I ran into was that cortex search seemed to return as much results as the limit given, which ends up returning a lot of irrelevant results. I ended up having to perform lots of separate cortex searches just to achieve a better result.
Accomplishments that we're proud of
What we learned
Prompt engineering makes a very huge difference when it comes to working with LLM and AI, a minor tweak in the prompt can make or break your app. Being able to articulate your words to get the model to do what you want if very important.
What's next for Recipe Helper
Right now the dataset I'm using only include some foods, which limits the accuracy. It would be good to add more food data. The dataset also includes a lot more nutrient values that's seen from the app, but for simplicity I've hard coded it to only show more common nutrient (carbohydrate, protein, fat), it would be good to allow users to also ask for other nutrient value (magnesium) if available.
Log in or sign up for Devpost to join the conversation.