←back to thread

Ask HN: Best way to create a searchable knowledge base?

20 points aljgz | 1 comments | 16 Oct 25 17:36 UTC | HN request time: 0.2s | source

Have you had the experience of using/developing knowledge bases? Here is my scenario:

My team is dealing with a lot of information: Wikis, Code repos, Monitoring dashboards, internal chat messages, emails, Task tickets, related systems, etc.

There are many cases when we need to do ad-hoc searches for anything related to a concept. For instance, imagine if someone makes a change to a metric, there is a need to find all dashboards that might be using this metric to make sure they are still valid after the change.

I don't want to just fix this problem, but create the ability to find related information in ad-hoc cases.

The ramp-up time is not important, as long as some positive value can be created with a small initial effort.

Any existing products (Paid/Free/Open Source, etc) and any references to existing knowledge (designs, discussions) about this would be really appreciated.

1. moomoo11 ◴[17 Oct 25 03:20 UTC] No.45613020[source]▶
I made something similar for a poc for an idea I was exploring.

Neo4j, open search, and vector embedding. I would use OpenAI api calls to generate the open search query based on user text input.

For example, user could search “what tasks are assigned to Jake that are at least 50% complete and due in the in the next 2 weeks” and it would be able to return relevant results.

Obviously only as good as the user search query. I spent close to a 100 hours writing tests to get it working close to 100% of the time. Eventually I dropped the embeddings because I could generate the opensearch query on the fly. So it was pretty lean and easy.