What this category's data looks like
This page addresses the vocational school student internship management scenario, covering three types of materials: internship agreements, internship plans, and internship logs. Internship agreements contain the names of both parties, internship position, start and end dates, signature and receipt fields; internship plans contain batch, number of positions, weekly schedule, and supervising teacher's signature; internship logs contain date, position, task entries, and signatures of the student and the on-site supervisor. The three types of materials come from the academic affairs and internship management roles, are updated by batch, and the version number and issuance date must be kept in sync. When the materials enter the knowledge base retrieval service, each serves as a source of candidate segments, to be recalled and compared by subsequent nodes or by humans when drafting signable material drafts and handover records, so as to avoid mismatches in names, quantities, and dates for the same record across two materials.
Configuration and implementation checks
This step is performed by the knowledge base retrieval service (retrieval node and search test), which outputs candidate segments and similarity scores (after pattern, merge, rerank, filter, and limit truncation). The signable material drafts and handover records are then completed by subsequent nodes or by humans.
| Object | Check | Verification action |
|---|---|---|
| Internship agreement | Check whether the names of both parties, position, start and end dates, signature and receipt fields are complete, and whether the version and issuance date are consistent | In the search test, use the agreement's key fields as the query, switch searchMode to compare the number of candidates and scores, and check whether the corresponding segments are recalled |
| Internship plan | Check the batch, number of positions, weekly schedule, and supervising teacher's signature; the quantity and dates must be consistent with the agreement | Test once each with synonymous questions, and observe whether the candidates after similarity filtering still retain the original text of the weekly schedule and quantity |
| Internship log | Check the date, position, task entries, and signatures of the student and the on-site supervisor; the date must correspond to the plan's weekly schedule | Test once each with usingReRank enabled and disabled, compare the ordering of the top few results, and confirm that the log date segments rank near the top |
| Retrieval return | The candidate segments and similarity scores constitute the referenced content; the number of references is recorded by the array length, and limit controls the upper bound | Check whether the returned candidates include both semantic and full-text results, record the array length separately, and do not compare it as equal to the referenced content budget |
Three common errors
- Symptom: The number of positions for the same internship record does not match between the agreement and the plan. Cause: The versions of the two materials are not synchronized, and retrieval hits segments from an old version without filtering by issuance date.
- Symptom: The log dates are misaligned with the plan's weekly schedule, and the handover record lacks a receipt. Cause: The signature and receipt fields were not updated along with the materials, and the recalled segments do not cover the latest version.
- Symptom: There are many candidates, but the ones ranked at the top are all irrelevant, and the answer lacks key paragraphs. Cause: The retrieval mode does not match the material form, rerank is not enabled, or similarity filtering is too broad.
How to verify the setup
Feature acceptance: Test once each with synonymously rewritten questions, compare candidates and scores, then toggle rerank on and off to observe the ordering of the top few results; switch searchMode to observe the difference between semantic and full-text results, and check the number of results after similarity filtering.
Business test: Draft one copy according to the template and complete one handover, and see whether missing items are blocked.
- In the knowledge base search test, switch
searchMode, record the number of candidates and scores separately, and confirm that both semantic and full-text results are merged. - Test once each with
usingReRankenabled and disabled, compare the ordering of the top few results, and separately record the number of reference array entries and the referenced text tokens.