Hybrid search / Qdrant

  • Resolved swimgeek

    (@swimgeek)


    3 days, 14 hours ago

    Hi, I’ve noticed the RAG store lookup process is not very good with finding keywords in our technical docs.

    I’m running Qdrant for RAG and I see it supports hybrid search: dense (semantic/vector) and sparse (lexical/keyword).

    Please let AI Engine chatbots use hybrid search.

Viewing 2 replies - 1 through 2 (of 2 total)
  • Plugin Author Jordy Meow

    (@tigroumeow)

    1 day ago

    You’re right, pure semantic search can miss exact keywords like part numbers or error codes, and hybrid search (dense + sparse with fusion) is the proper answer for that. It requires changes in how AI Engine indexes content with Qdrant, not just how it queries, so it’s not a small addition, but it’s on my radar 😊

    Meanwhile, two things can help with recall: increase the max results and lower the min score in your embeddings environment settings, and make sure important keywords appear in the chunk content. Also, since Qdrant support is part of the Pro version, feel free to contact me through Meow Apps website (https://meowapps.com/contact) so I can follow up on this properly.

    Thread Starter swimgeek

    (@swimgeek)

    4 hours, 3 minutes ago

    Hi

    My min score is currently very low to try work around this issue: 30.

    OK, thanks, will open a ticket with support.

Viewing 2 replies - 1 through 2 (of 2 total)

You must be logged in to reply to this topic.