UniCopy

Inspiration

We all want to have the next billion-dollar idea - but how can we ensure it's our idea? Presenting UniCopy, the one-stop chatbot for making sure you don't get sued while pitching your unicorn.

What it does

The user enters their idea into UniCopy. We then search our vector database for registered patents with similar ideas, providing the user with insight into competitors and headwinds and the patent number itself so they can conduct further research into existing ideas which share similar functionality or solve a similar problem.

How we built it

We used Cohere's API for constructing the Retrieval-Augmented Generation (RAG) model, including embedding, storing, and retrieving documents, as well as generating the output. We used LangChain to efficiently and smartly chunk the text of our documents to ensure the semantics of each chunk remain unchanged while reducing latency for output generation. We then created a React frontend with a Flask backend to create a GUI.

Challenges we ran into

Creating the RAG and ensuring the documents retrieved and output generated are accurate and relevant to the user query; creating the documents to be embedded from a BigQuery storage location.

What we learned

How to create RAGs with custom embeddings and how to prompt engineer the model to achieve our needs.

What's next for UniCopy

Checking if UniCopy is original :D

Built With

• cohere
• langchain