What data for this category looks like
This category covers full-process documents for manufacturers of single-use consumables, implantable and interventional devices, diagnostic equipment, and active medical devices. Documents include production process specifications, batch production records, clean zone records, sterilization batch records, inspection reports, software version records, adverse event and traceability materials. Data sources include design history files, device and material batches, MES systems, clean environment monitoring equipment, sterilization equipment, and inspection and calibration systems. Updates occur once per batch or per device, with regular synchronization per environmental monitoring plans. Data fields include unique identifiers for critical materials, sterilization batch numbers, clean environment parameters, software version numbers, and other fields. Some documents require association with batch and supplier information.
Constraints on knowledge base retrieval and recall
Data for this category is multi-source, heterogeneous, and updated frequently. Retrieval systems must support multi-source data access and incremental synchronization to prevent audit deviations caused by data lag. Document types include process specifications, production records, traceability materials, and other categories. Retrieval workflows must adapt chunking and retrieval strategies for different document types to ensure no key information is missed. Data fields include unique identifiers for critical materials and batch association information. Retrieval must accurately match data linked to specific batches and suppliers to avoid recalling cross-batch or unrelated documents. Audits involve aseptic control and traceability requirements. The accuracy of retrieval results directly impacts product compliance, so retrieval matching accuracy must be strictly controlled.
Configuration settings
The following configuration settings are based on the functional logic of FastGPT v4.17.0.
Choose a retrieval mode through representative-query testing; configure category, batch, or entity filters separately because the mode itself does not enforce those boundaries.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Medical device audits require high matching accuracy to prevent unrelated clean zone records or non-target batch documents from appearing in retrieval results |
limit | Top 3–5 results | Single batch production records have compact content. Too many results will exceed the context window and reduce the coherence of audit logic |
usingReRank | Enabled | Low-match sterilization batch records and environment monitoring data must be filtered to improve recall accuracy for key process and material identification information |
datasetSearchUsingExtensionQuery | Disabled | Medical device audits require strict matching of original document content to prevent extended queries from introducing non-compliant unrelated information |
collectionIds | Bind dedicated knowledge base collections for clean production and supplier audit | Isolate documents from different categories and batches to prevent cross-category or cross-batch data from interfering with retrieval results |
Three common mistakes
- Phenomenon: Non-current batch sterilization batch records appear in retrieval results. Cause: The
collectionIdsconfiguration to limit dedicated knowledge base collections is not set, allowing cross-batch data to be recalled. - Phenomenon: Knowledge base search input and response reference content are displayed at the end of the answer. Cause: Reference display configuration is not disabled in the workflow node, or post-retrieval output filtering rules are not set correctly.
- Phenomenon: Retrieval returns 0 results. Cause: The
similaritythreshold is set too high, causing eligible batch production records to not be recalled, or no matching batch document data exists.
How to verify correct configuration
- Upload a single typical batch production record and clean zone record, run a retrieval test, and confirm the results only include documents from the target category.
- Trigger a workflow call, and check that no knowledge base search input and response reference content are displayed in the answer.
- Adjust the
similaritythreshold, and verify that the matching accuracy of retrieval results meets the precision requirements of business audits. - Check the locally deployed knowledge base storage configuration, and confirm that the number of uploadable files meets the storage needs of multi-batch documents.
The configuration values provided are common starting points, and should be evaluated against the reader's own samples.