What this category’s data includes
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 documents cover production process specifications, batch production records, clean zone operation records, sterilization batch records, inspection reports, software version records, adverse event and traceability materials. Data is updated upon completion of each product batch or device, or according to the clean environment monitoring plan. Documents include fields such as device unique identifiers, material batch numbers, sterilization parameters, clean zone temperature and humidity records, and software version numbers. Units match the corresponding metrology requirements based on document type.
What constraints this category’s characteristics impose on knowledge base retrieval and recall
This category’s data sources are scattered and linked to batches and unique identifiers. Retrieval must precisely match device IDs or batch numbers to avoid cross-batch confusion. Update frequency changes with production batches. Retrieval must support real-time or near-real-time recall of the latest batch data. Document types are diverse and have large field differences. This requires support for mixed retrieval across multiple document types, while retaining field-level precise matching capabilities. Business error consequences involve patient safety and compliance risks. Retrieval result confidence levels must match the risk level, and irrelevant or expired data must be avoided in recall results.
Configuration Settings
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Match the precision requirements for medical device-specific fields, and avoid recalling low-relevance general production documents |
limit | Top 6–8 entries | Control the number of single-batch retrieval results, and adapt to the short drafting requirements of process documents and abnormal reports |
searchMode | mixedRecall | Balance semantic matching and precise retrieval of keywords such as device unique identifiers and batch numbers |
usingReRank | Enabled | Reorder recall results by business field relevance, and prioritize returning core documents such as sterilization parameters and clean zone records |
datasetSearchUsingExtensionQuery | Enabled | Supplement retrieval for associated fields such as device models and software versions, and cover extended query needs during drafting |
collectionIds | Bind exclusive datasets by business scenario | Isolate production data across different product lines, and avoid confusion across category documents |
Three common misconfigurations
- Symptom: Retrieval results do not match the specified device’s process specification documents, and only return general production content. Cause: Keyword retrieval mode is not enabled, and precise matching for fields such as device unique identifiers and batch numbers is not configured.
- Symptom: Automatic addition of Bearer tokens when external data source APIs are called causes requests to be rejected by the data source. Cause: FastGPT’s automatic token supplement function for API requests is not disabled, which does not comply with the data source’s authentication requirements.
- Symptom: Recall results include expired sterilization batch records, leading to inaccurate abnormal report content. Cause: Exclusive datasets are not bound to isolate production data across different batches, and recall results are not filtered by update time.
How to Confirm Configurations Are Correct
- The FastGPT knowledge base management page may be accessed to check if
collectionIdsis bound to the exclusive dataset for the current product line. - A test query may be initiated, with a device unique identifier or batch number entered, to verify that corresponding production documents can be precisely recalled.
- The retrieval result list may be viewed to confirm that
usingReRankis active, and core documents such as sterilization parameters and clean zone records appear at the top. - The knowledge base API interface may be called to check if the request header includes additional automatically supplemented Bearer tokens, or to confirm that the token supplement logic has been disabled.
The values provided are common starting points and should be measured against the reader's own samples.