let me think through it.

One way could be to have an LLM review the full discussion and output a file with predefined metrics (like collaboration vs conflict signals could be one). We could store that and pull it into the dashboard. Since these patterns don't change often, running it every 3-6 months would probably be enough.

that said, there might be a smarter, cheaper, and deterministic approach using proxy variables. I'll keep thinking on it.

sorukumar

I’m wondering how the work relationship of people could be discovered. Some people reply to each other because they are collaborating. Some people serially reply to each other, because they cannot ever agree. If the latter could be discovered in some manner, and push people apart rather than pulling them together, the clusters may actually be more representative of what’s going on.