How does our search index work?

How the search index works

Vector Embeddings in Search

Vector embeddings are at the core of our search system. They allow us to convert both documents and search queries into numerical representations that can be compared efficiently. This results in more accurate and relevant search results. Read more about this on Vector calculations.

Fast and Accurate Search Results

Our search system uses algorithms to find the most relevant documents based on your query. This ensures that you quickly gain access to the most relevant information, even when your search terms don't exactly match the words in the documents.

1. Document indexing and summarization: All documents in our database are indexed and summarized. This involves several steps:

  • Dividing documents into smaller parts (chunks).
  • Converting chunks into numerical representations (embeddings).
  • Placing embeddings in a vector database for fast and efficient searching.
  • Generating brief summaries for each case from sources such as the Court Database, the National Tax Tribunal, the EU Court of Justice, etc. This provides a quick overview of the case content.

This process ensures that we can both search efficiently and provide users with quick insights into the content of the cases.

How document chat works

2. Processing search query: When you enter a search query, it is also converted into an embedding. This allows your query to be compared with the indexed documents in a more sophisticated way than traditional text search.

How document chat works

3. Retrieving relevant documents: The system finds the documents or chunks whose embeddings are closest to your querys embedding. This ensures that the most relevant results are returned, even when they dont contain exactly the same words as your search.

How document chat works

4. Presenting results: The found documents or chunks are ranked by relevance and presented to you. Each result typically contains an excerpt or summary of the relevant content, allowing you to quickly assess if its what youre looking for.

How document chat works