What the data for this category looks like
Data for the warehousing, lot tracking and in-plant logistics category originates from ERP/WMS systems, barcode/RFID equipment, purchase work orders, storage location management, AGV scheduling systems and weighing equipment. Daily documents include receiving slips, putaway slips, material picking and return slips, warehouse transfer slips, inventory counts, lot ledgers, first-in first-out records and expiration date records. Fields cover material codes, lot numbers, storage location numbers, expiration dates, quantities (units: pieces, pallets or kilograms) and other relevant items. Data update frequencies are split into three categories: real-time (such as AGV scheduling data and weighing data), daily (such as material picking and return records) and monthly (such as inventory counts).
What constraints these characteristics impose on knowledge base retrieval and recall
Multi-source heterogeneous data sources require retrieval to cover document formats across different systems, and balance matching for both structured fields and unstructured text. Documents with varying update frequencies require a hybrid strategy of incremental synchronization and full updates to prevent retrieval of expired data. Strongly related business fields such as lots and storage locations require retrieval to support multi-condition precise matching, and cannot rely solely on general semantic retrieval. Cross-process material flow management requires retrieval to associate lot ledgers with logistics documents, covering the complete business chain.
The following configurations are based on FastGPT v4.17.0.
How to set the configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Match strongly related business fields such as material codes and lot numbers, filter irrelevant results with low similarity |
limit | Top 10 entries | Single recall covers multiple documents and ledger data required for work order completeness checks and inventory-book vs physical inventory verification |
searchMode | mixedRecall | Adapt to structured and unstructured documents from multiple sources, balance semantic understanding and precise field matching |
usingReRank | Enabled, re-rank top 5 entries | Prioritize display of matching results for core business fields such as expiration dates and storage locations, improve business review efficiency |
datasetSearchUsingExtensionQuery | Enabled | Automatically expand retrieval scope to cover associated data of lot ledgers and logistics documents |
collectionIds | Specify warehouse and lot-related knowledge base IDs by business domain | Limit retrieval scope to avoid interference from irrelevant documents during business review |
Three common configuration mistakes
- Retrieval results include non-warehousing domain documents, and external release channels cannot properly reference results. The cause is failure to correctly specify the
collectionIdsparameter, or failure to bind corresponding knowledge base permissions in the release channel's API configuration. - Retrieval results include paragraph ID fields, which interfere with display of core business information such as lots and storage locations. The cause is failure to turn off the paragraph ID output switch, and failure to configure field filtering for structured business documents.
- Retrieval fails to cover cross-process material flow data, such as failure to recall both lot ledgers and AGV scheduling records. The cause is failure to enable
datasetSearchUsingExtensionQuery, or only configuring pure vector retrieval mode without enabling keyword matching.
How to verify a correct configuration
- Manually enter a query term combining a material code and lot number, check that returned results only include warehousing and lot-related documents with no irrelevant content.
- Review the returned fields of retrieval results, confirm that non-business necessary information such as paragraph IDs is not included.
- Simulate a work order completeness scenario, enter a query term including a work order number and material requirements, check that the number of returned results meets business verification needs.
- Test external channel API calls, confirm that returned results can be properly parsed and displayed.
The values provided are common starting points, and should be evaluated against the reader's own samples.