What Data for This Category Looks Like
Typical entities in this category are flow, pressure, temperature, and analytical instrument manufacturers and metrology laboratories. Daily documents include assembly and commissioning sheets, calibration procedures, original calibration records, error curves, metrology certificates, and software version logs. Data sources cover design documents, standard instrument and environmental monitoring data, calibration software outputs, laboratory information system records, and customer range requirements. Data is updated after each calibration of every instrument. Documents contain precise numerical fields and units, such as range limits, error values, traceability numbers, and calibration cycles.
Constraints on Knowledge Base Retrieval and Recall
The multi-source nature, high-frequency updates, and precise numerical attributes of data in this category require retrieval to accurately match fields such as the customer-specified range and error threshold, to avoid recalling irrelevant documents. High-frequency updated calibration records must be linked in real time to the bidding requirements of the corresponding instrument. The retrieval scope must be limited to the set of instruments involved in the current bid. Precise numerical documents with units cannot be matched via general keywords. A combination of vector retrieval for numerical fields and keyword retrieval is needed to ensure the accuracy and traceability of recalled content.
Configuration Settings
The following configurations are formulated based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Accurate matching of precise fields such as ranges and error thresholds in calibration procedures and metrology certificates is required to avoid recalling irrelevant general documents |
limit | Top 8 entries | Bid documents need to cover calibration requirements for multiple instruments. Too many results will exceed the context length limit, while too few will miss critical parameters |
searchMode | mixedRecall | Balances vector retrieval for numerical calibration parameters and keyword retrieval for fields such as traceability numbers and standard instrument models |
usingReRank | Enabled | Calibration-related documents have similar structures. Reranking prioritizes recalling results that include the customer-specified range and error requirements |
datasetSearchUsingExtensionQuery | Enabled | Bidding requirements often include implicit customer range extension requirements. Extended queries can recall applicable calibration procedures |
collectionIds | Specify the ID of the calibration-related document collection | Avoid mixing documents from other categories to interfere with retrieval results, and only cover calibration and certificate documents required for the current bid |
Three Common Mistakes
- Phenomenon: After uploading an original calibration record file, the file content is not recalled during retrieval. Cause: The uploaded file was not associated with the specified
collectionIdscollection, so the retrieval scope does not cover the collection to which the document belongs. - Phenomenon: A request failure prompt is returned when calling the interface to add content to the knowledge base. Cause: Internal network HTTPS access interface forwarding rules were not properly configured via nginx reverse proxy, so interface requests cannot reach the server normally.
- Phenomenon: The matching degree of retrieval results for tabular error curve data is very low. Cause: Multi-vector support for tabular data was not enabled, so numerical fields in the table were not properly split and vectorized, and cannot match the range and error threshold keywords in the retrieval.
How to Confirm Proper Configuration
- Upload a calibration procedure file, and check if the file is automatically associated with the specified knowledge base collection. This can be confirmed via the file list in the knowledge base management interface.
- Enter a search term that includes the customer-specified range, and check if retrieval results preferentially return calibration documents for the corresponding range. This can be confirmed via the similarity score of the retrieval results.
- Upload a tabular error curve file, and check if the system can extract numerical fields from the table for matching. This can be confirmed via previewing content snippets of the retrieval results.
- Call the interface to add a calibration record, and check if the record appears in the knowledge base content list. This can be confirmed via the content statistics in the knowledge base management interface.
The values given are common starting points and should be measured against the reader's own samples.