GenChef

Sample Generation

Inspiration

We're all software engineers, so we are prone to try and find the most efficient way to accomplish chores, and what chore do we have to do more often than preparing meals?

What it does

GenChef utilizes a retrieval augmented large language model and image generation model backed by over 450,000 real recipes in a semantic vector database to help provide ground-truth results for delicious and mindful recipes.

How we built it

Weaviate for the Vector Database LllamaIndex for the LLM-Database Interface LangChain for managing Semantic Embeddings for Query ChatGPT-3.5 for the LLM Ada-002 for the Semantic Embeddings Model Streamlit for the frontend

Challenges we ran into

Vectorizing 450K recipes in a NoSQL vector database took a long time... and a lot of hassle with the APIs.

Accomplishments that we're proud of

Our model has the ability to reference ground-truth recipes across a wide array of cuisine and even reference its sources.

What we learned

Retrieval augmented generation is an excellent way to make large language models more useful at home and in the workplace as it significantly decreases the incidence of hallucinations.