Automatic unstructured PDF ingestion on GCP

The scope of this session is to show how it is possible to create a Question Answering agent on GCP starting from complex documents, like technical handbooks. This will be done by using the following passages:

1. Document Segmentation
Semantic structures (like title, sections, tables...) will be inferred with State-of-the-Art models.

2. Optical Character Recognition (OCR)
Google OCR will be applied to the previously-determined structures in order to determine their content.

3. Vectorization
The text will be saved in a vector store in order to allow a fast information retrieval.

4. Large Language Model (LLM) Connection
The vector store will be connected to an LLM that will be used to provide the user with suitable answers for his/her questions.

This project will showcase the capabilities of State-of-the-Art models and technologies in the field of Generative AI.

Speaker Bio:

Stefano Cariddi earned his Ph.D. in Astronomy in 2018 and then transitioned to the field of Artificial Intelligence. His profile is that of a Data Scientist, so he possesses strong skills in data processing, dataset creation, model training, and, more broadly, Machine Learning. His profile is further complemented by 10 years of experience in scientific programming in Python, the use of pre-trained models in Computer Vision, Natural Language Processing, and Text Generation, and the utilization of the Google Cloud Platform for delivering Artificial Intelligence services.