A user asks questions, top k related wikipedia articles are chosen (NLP model) and questions are answered based on each of those articles (NLP model).
Video demo: https://youtu.be/A4AFAbyFQNw
- Question prompt -> Model to correct text & complete - correct.py - ref to complete phrase -> Model detect most relevant article [V] - sentence-transformer (https://www.sbert.net/examples/applications/semantic-search/README.html) -> Model answer question in that article - distilbert / longformer -> ask if answer was useful -> if not, answer on 2nd most relevant article -> UI with Gradio (https://www.gradio.app/)