What the data for this category looks like
Core data for this category comes from design history files, device and material batch ledgers, MES systems, clean environment monitoring equipment, sterilization equipment, and inspection and calibration systems. Daily generated documents include production process specifications, batch production records, clean zone operation records, sterilization batch records, inspection reports, software version logs, adverse event and traceability materials. Each document includes controlled version numbers, batch identifiers, and key process parameter fields. Documents are updated for each product batch or each piece of equipment output. Clean environment data is synchronized regularly per established monitoring plans.
What constraints these characteristics impose on knowledge base retrieval and recall
Documents in this category are mostly controlled professional records tied to specific batches. Data sources are scattered across multiple systems. Retrieval must accurately match batch identifiers and key process fields. Frequently updated batch data and regularly synchronized clean environment data require the retrieval system to support filtering recall results by time range and batch dimension. Business weights vary significantly across document types. Recall priority for sterilization batch records and traceability materials must be higher than for general operation records. Expired version documents must be excluded from recall. Errors directly impact patient safety. Retrieval and recall results must cover all controlled documents related to the current assembly and inspection processes. No critical validation basis may be omitted.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Documents in this category have high professional complexity. This range filters low-correlation results and avoids recalling irrelevant batch records |
limit | Top 10 entries | This ensures all relevant documents for the current assembly process are covered, while controlling context length to stay within model limits |
usingReRank | Enabled | This reorders recall results, raising the ranking priority of documents related to key business fields such as sterilization parameters and batch identifiers |
datasetSearchUsingExtensionQuery | Enabled | This supplements retrieval for exclusive fields such as batch identifiers and process parameters, improving recall accuracy for documents in this category |
collectionIds | Bind to the exclusive dataset ID of the current production line | This isolates batch data across different production lines, preventing recall of controlled documents from unrelated production lines |
Three Common Misconfigurations
- Symptom: A code 514 error is returned when calling the knowledge base retrieval API. Cause: The
collectionIdsparameter is not configured correctly, and a dataset ID from a non-target production line is bound, triggering a permission check failure. - Symptom: Expired version production records appear in retrieval results. Cause: Document version-based filtering logic is not enabled, and extended queries are not used to match currently valid controlled documents.
- Symptom: The number of retrieved and recalled results is significantly lower than the actual number of relevant documents. Cause: The
similaritythreshold is set too high, filtering out documents that match professional terminology but meet business requirements, or thelimitparameter value is too low.
How to Confirm Proper Configuration
- For FastGPT v4.17.0, call the retrieval API, pass the current batch's identifier as the search keyword, and verify that the returned results include core documents such as the batch production record and sterilization batch record for this batch.
- Adjust the
similarityparameter value, and confirm through multiple rounds of testing that the relevance of returned results meets business review requirements. - Verify that when
usingReRankis enabled, documents related to key process parameters appear at the top of the results. - Check that the
collectionIdsparameter is bound to the exclusive dataset for the current production line, to avoid recalling irrelevant data from other production lines.
The values provided here are common starting points, and should be evaluated against the reader's own samples.