-
Story
-
Resolution: Done
-
Normal
-
None
-
None
-
None
-
False
-
-
False
-
None
-
Unset
-
None
-
-
The idea here is to expose a generic summarization API that can essentially do "RAG on the fly" with any provided context and system prompt. This would allow for agents to use it to summarize data from different sources. The summarization API is part of it, the other is an agentic part. Here we would have agents that are essentially API endpoints that know a data source to query and have a uniqe system prompt. We could have one for app sre changelog, one for webrca, whatever. The final part is agent routing. We can do this via NLP in the same way we do quality detection and junk removal. We'd train an in memory model on known good prompts.
The flow would be like this:
- User types in "When will my merge request $URL going to merge?"
- Agent routing detector determines this is a request to look at the merge queue
- The request is routed to the merge queue agent
- the merge queue agent first queries the changelog to get the required content
- the LLM is prompted with the merge queue agent system prompt and the data
- The answer is generated for the user
There are no Sub-Tasks for this issue.