AI for document generation

Project description

Get answers to everything you wondered about! We have made a chat robot where you can ask different types of questions on documents and then get answers fairly quickly with the help of AI language technology. It can be about:

- Questions in the form of free text "What does it cost to build the Norrbottnia line?" or "What are the benefits of it?"

- Find out the places, people, times, measurements or organizations involved

and / or

- Get a shorter summary of the most important things in a long text with specially selected sentences.

What is unique is that we have not entered any more information than the raw documents (in pdf, downloaded from web pages) that the computer analyzes on its own and that the response times are only about 20 seconds for 300 pages thanks to smart use of computational graphics cards of a new powergul model, a standard fast computer would have had to think like an hour to thresh through these pages and cloud services / supercomputers might have boot/start-up times of several minutes or sometimes weeks.

The technology is called Natural Language Processing, NLP, and thanks to efforts from the Royal Library (KB), there are now also opportunities to handle documents in Swedish. The AI ​​model (BERT) is originally from Google and the PyTorch programming library is from Facebook's AI lab.

We have made a prototype with a publicly accessible server that has been on duty for a few weeks for tests of the question mechanisms and in it we have included documents about the Norrbottnia line (typically environmental impact assessments), Ostlänken and some older books (Strindberg's Red room and Lagerlöf's Nils Holgersson's wonderful journey through Sweden).

Test!

Here you can test drive the chat bot

Background information