Xaperio
Ask technical questions and get answers drawn directly from open books.
Xaperio is an extractive question-answering system built around openly licensed technical books.
Instead of just searching for exact keywords, it uses semantic search to understand the meaning of a question and find sections of books that are most likely to contain the answer.
The contents of the books are split into sections, converted into vector embeddings, and stored in PostgreSQL using pgvector. When a user asks a question, Xaperio searches through these sections for the most relevant ones. An extractive question-answering model then looks through the retrieved text and picks out the part that directly answers the question.
The answer is not generated by AI. It comes directly from the source book, and the user can also see the section it was taken from for more context.
For example, someone could search:
“How does TCP congestion control work?”
and get an answer from relevant sections of books such as:
-
Computer Networks: A Systems Approach
Chapter 3: Transport Layer
Section 3.4: Congestion Control -
Beej’s Guide to Network Programming
TCP/IP
Flow and Congestion
I’m still working on improving how well Xaperio finds the right sections and how accurately it extracts the final answer.