HomeAboutProjectsWritingResume
← Projects

What I did

Patent Analyzer

AI patent document analysis with large language models and hybrid search

EcoPatent Analyzer is a patent analysis platform. It scrapes patent data from the USPTO website. It classifies inventions with the 40 TRIZ principles. A chatbot answers questions with hybrid search in a Qdrant vector database. The system reranks the results with Cohere. It stores patent data in Azure Cosmos DB and Azure Blob Storage.

PythonDjangoQdrantDBRAGCohereReactDocker
View on GitHub →

Features

  • The system scrapes the full patent text from the USPTO website.
  • It classifies the patent with the 40 TRIZ principles.
  • It runs a chatbot with hybrid dense and sparse search.
  • It reranks the search results with Cohere.
  • It generates a PDF report of the analysis.
  • It stores patent data in Azure Cosmos DB and Azure Blob Storage.

How it works

  1. Upload a patent PDF.
  2. Let the system scrape the full patent text.
  3. Let the system classify the patent with TRIZ.
  4. Ask a question in the chatbot.
  5. Read the answer and the PDF report.

Tech stack

  • Backend: Python 3.13, Django 6, Django REST Framework
  • Frontend: React 18, TypeScript, Vite, Tailwind CSS
  • Models: OpenAI GPT-4o, TogetherAI DeepSeek
  • Vector store: Qdrant Cloud
  • Reranker: Cohere rerank-english-v3.0
  • Storage: Azure Cosmos DB, Azure Blob Storage
  • Scraping: Selenium, BeautifulSoup