This simple solution uses OpenAI's API for both embedding and generation. I use a windowing approach (2-page chunks with 1-page overlap) to maintain a relatively low number of requests to the embeddings endpoint while accommodating cases in which a relevant passage is split across two consecutive pages. Due to the scale and nature of the assignment (no more than 1000 documents per session, no need for persistence), I forgo the use of sophisticated vector databases & NNS implementations in favor of an in-memory embedding matrix & complete similarity search. A similar system with greater performance requirements should probably use an optimized vector database with smaller, sensible chunks (perhaps with overlap) for fast and precise lookup. Such a system should also implement guardrails to reduce hallucination and adversarial attacks.
The main script, chatbot.py, generally leaves error messages unexposed to the user, instead printing generic error messages such as "An error occurred while searching the PDF." In a higher-stakes setting, where bugs needed to be identified and fixes rolled out, I would probably configure the script to report error tracebacks to some central logging service, accessible only to the developers.
This project has only been tested with Python 3.9. To set up your environment (e.g., conda, venv), run:
pip install -r requirements.txt
Note that if this project were to grow any larger and more complex, I would probably migrate it to uv.
$ python chatbot.py USHistory-WEB-chapter1.pdf
Ingested USHistory-WEB-chapter1.pdf. Please ask your questions.
>>> How did slavery in Africa differ from slavery in the Americas?
Slavery in Africa was generally not permanent or inherited; children of enslaved people were usually free and could intermarry with their captors. In contrast, slavery in the Americas became permanent and race-based, with enslaved people identified by their skin color, and the children born to enslaved people automatically becoming enslaved. This change was driven by the demand for a permanent, plentiful labor supply for labor-intensive crops and shaped the unique character of slavery in the New World.
I found this answer somewhere on page 20 and/or 21.
>>> What was the religious significance of gold in the Incan Empire?
The Inca worshipped the sun god Inti and called gold the “sweat” of the sun.
I found this answer somewhere on page 7 and/or 8.
>>> Why do extreme trans-Neptunian objects have elongated orbits?
I can't find the answer to your question. Sorry!
>>> exit
PDF courtesy of OpenStax