Created a Japanese PDF RAG demo that shows 'what was seen to answer' rather than 'likely-sounding answers'
Matsunobe, a student at ZEN University, developed a Japanese PDF RAG demo that displays citations (file name and page number) to clarify what the model looked at to answer.
Matsunobe, a second-year student in the correspondence course at ZEN University, is studying by building LLM-related mechanisms to become an AI engineer. Matsunobe noted that while LLMs typically provide answers in plausible Japanese when asked about materials, there is a concern regarding where in the materials the model is looking to provide that answer.
To address this, a RAG demo was created that allows users to upload a PDF and ask questions in Japanese, providing answers along with citations including the file name and page number to clarify what the model consulted.
This item is a summary. The full, checkable version is at the original source:
Read the original at Zenn (LLM) ↗