AI Interview Copilot/Web Experience/Knowledge base issues

Knowledge base issues

Explain knowledge-base upload, retrieval hits, and performance issues based on frontend/backend parameters and processing flow.

Preconditions for taking effect

  • Answer requests must explicitly carry useKnowledgeBase=true.
  • The number of participating documents is controlled by knowledgeBaseMaxDocuments, with a range of 1-10.
  • A toggle being on does not mean retrieval will definitely take effect. There must still be valid documents and a related question.

Upload constraints

  • Single file maximum: 50MB.
  • Total document count maximum: 10.
  • Single document character limit: 200000.
  • Parse failure or index failure interrupts the upload flow and does not enter the retrieval-hit stage.

Frequent issue diagnosis

  • Upload failed but the toggle can be turned on: usually the upload stage was blocked by a limit, not a retrieval failure.
  • Answer does not change after updating resume: first rule out that the question context did not change, then run an on/off comparison test.
  • Response becomes slower after enabling: usually expected cost from retrieval and context assembly.

Standard troubleshooting steps

  1. Check whether upload succeeded: size, count, character length, and quota.
  2. Check whether indexing completed: whether the document remains in the list and is retrievable.
  3. Check whether request parameters are passed correctly: useKnowledgeBase and document count limit.
  4. Finally, run an A/B test with the same question: knowledge base off vs on.

Effect acceptance criteria

  • Answers show an explainable difference when the knowledge base is off vs on.
  • Document-related questions can consistently hit the corresponding material.
  • The more focused the documents are, the higher the answer relevance.

Slower response after enabling the knowledge base is usually expected, because retrieval and context assembly are added to the answer chain.