PatentAI Analyser

The Software analyzes the novelty of your invention ideas against a database of existing patents from Google Cloud. It leverages vector embeddings and a chat interface to help innovators.

Comment

Inspiration

We wanted to simplify the first step of innovation: finding out if an idea already exists. Standard patent databases are difficult to search with their legal jargon and keyword-based systems. We were inspired to build a more intuitive tool that lets anyone check their ideas using plain English.

What it does

PatentAI Analyst is a web application that helps users find existing patents that are conceptually similar to their own invention idea. A user describes their idea in a chat-like interface. The app analyzes the text and uses MongoDB's vector search to retrieve the most similar patents from its database. It then displays these results and allows the user to see a detailed, side-by-side comparison between their idea and any found patent.

How we built it

We used a modern, efficient technology stack. Backend: A Python API built with FastAPI. Database: MongoDB Atlas, used as our primary vector database for storing and searching patent data with its native Vector Search capabilities. AI Model: Sentence-Transformers, an open-source library, was used to convert patent abstracts into vector embeddings for semantic search. Frontend: A dynamic single-page application built with React, using Framer Motion for UI animations.

Challenges we ran into

Our main challenge was a data issue where our vector search was failing silently. We discovered a subtle data type mismatch between the embeddings our script generated and what the Atlas Search index expected. We solved this by re-ingesting our data with a sanitized, correct data type. This taught us to always verify our data pipeline at every stage.

Accomplishments that we're proud of

We are proud to have built a working, full-stack application that solves a real problem. The core vector search functionality works correctly, providing relevant results from a natural language query. We are also pleased with the polished and interactive conversational UI, which makes the complex process of patent searching feel simple and accessible.

What we learned

This project provided deep, hands-on experience with AI-powered search. Our biggest takeaway was how capable MongoDB Atlas is as an all-in-one document and vector database. Its native Vector Search feature allowed us to build our core functionality without adding a separate, specialized database service, greatly simplifying our architecture.

What's next for PatentAI Analyst

There are three clear paths for improvement: Activate Cloud AI: Integrate Google's Gemini Pro API to provide much more intelligent and nuanced comparisons, moving beyond the current templated analysis. Expand the Dataset: Ingest millions of international patents to provide global search coverage. Add User Features: Implement user accounts to allow for saving search history and tracking patents of interest.

Built With

Share this project:

Updates