The Semantic Finder

Semantic Finder enables engineers to find the desired information in technical documentation efficiently and independently of specific search terms (search for the meaning of a phrase). This is achieved through a full-text and semantic search (enabled by Transformer language models) in various types of indexed documents or other text data sources. Users can get answers to their questions based on the search results (Question Answering). During the project, we conducted an evaluation of the involved search methods and language models with engineers on technical document datasets they are familiar with. The language models evaluated include pre-trained transformer models, domain document customized transformer models, sentence transformer models and multilingual models.