Open
Description
Mining knowledge-base entries from GitHub issues ("question" label). The answer from a ClickHouse employee is usually authoritative. We need to use LLM to decide if the answer provides a solution and to convert to Q&A.
https://clickhouse-inc.slack.com/archives/C03A9B0H05T/p1750092455249029
Note: we are already using ClickHouse/ClickHouse issues as a source for the Ask AI widget. However the number of times that it uses Github issues to answer a question is very low. Unfortunately there doesn't seem to be much control over how it uses this source, or even any insight into how it chooses these issues. Some custom solution for this task is an interesting option to explore so that we can add the answers as questions.