Document current implementation in gherkin

[bwilczek]

Main use cases to be covered (from pairing session with with @id-ilych)

1. Populate knowledge base. Use CLI -> fetch from Confluence (pages with all the comments) -> SQL -> OpenAI generate embeddings -> vector db
2. Q&A. User posts a message to slack channel the bot is invited to. Slack listener receives the event -> detects if it is a question (if it has ? at the end) -> generates embedding for the question (using OpenAI) -> retrieves most relevant documents from vector database -> builds a message that includes content of relevant pages + user’s question -> creates a conversation with Open AI (thread in an assistant) -> wait for a response -> post to Slack as a response (in a thread) -> store it in database (including association between assistant thread id and slack thread id)
3. Feedback. User posts a message to a thread started from the original question -> same as Q&A but instead of creating new conversation it posts to an exiting one.
4. Identify knowledge gaps. Use CLI -> initiate the process by selecting a topic -> it then uses embeddings and OpenAI to select unanswered questions about that topic -> unanswered questions are posted to specific slack channel -> users can answer them in threads -> bookmarked conversations are uploaded to Confluence (dedicated space).
5. Gamification (cross-concern). Asking questions rewarded with seeker points. When you question was unanswered you get revealer points during knowledge gap discovery. Luminary points are given to those who answer knowledge gap questions

[id-ilych]

Documents are always found (at least with the current implementation, though maybe we could set a distance threshold. So it is up to OpenAI to recognize that given documents don't have info about that.

[id-ilych]

Revealer points are rewarded only when knowlegde gap search is initiated. At least with the current implementation.

[id-ilych]

Just in case - vector db stores page ids, not titles. Though I think we could build a matcher that will transform it automatically.