This was a test of how an integration of a chatbot focused on privacy could be integrated into an LMS like Canvas.
The initial purpose of this was to test how a privacy-focused chatbot could ingest course materials to help students with basic questions on course content.
Warning
This relies on a specific branch of privateGPT, released in May 2023, known now as the "primordial" version. That project has since moved on and includes its own web interface. I built this around the same time to see what was possible to integrate into the Canvas Learning Management System. See the Further Reading section for the write-up on this project.
Important
To get this running (for whatever reason), you will need to install the primordial version of privateGPT, and then all of the requirements from the requirements.txt
file. This will not work without those prerequisites!
- Add an appropriate AI model to the
models
directory. - Add source documents to the
source_documents
directory and runingest.py
to give privateGPT access to your documents. - Watch for any errors - at this stage it is likely missing or conflicting requirements.
Once privateGPT is installed, the app.py
file and template
directory can be dropped into the same folder. Running the app.py
file should launch privateGPT and the web server.
The Course Helper uses a custom made "bridging app" that launches privateGPT and a webserver, then connects the two.
Once created, the web interface portion, running locally, allows us to plug it into a Canvas page, like so:
This video is a sped up version of how the bot works.
Chatbot0_ecampus_essentials.mp4
Main article and write-up at the OSU Inspire Blog: You won’t believe where AI could replace you next! (But you’re probably going to like it)