Uses TypeChat to analyze original messages and save the analysis to the scrubbed_messages collection before the original messages disappear
Refactors our TypeChat usage to move it into a more general use case chat-core
Notes
This tries its best to scrub PII, but there's a possibility that PII will still come through. I'm thinking that when we look at the analysis in aggregate, some of the topics or sentiments that might indicate PII will stand out and we can simply delete those entries. I also added a PII flag and in my testing ChatGPT is pretty good at detecting whether the message contains PII in the first place.
Jira: none
Changes
Notes