What Data for This Category Looks Like
This category of data primarily comes from ERP/WMS systems, barcode or RFID collection devices, purchase work orders, storage location management ledgers, AGV transport 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 (FIFO) and expiration date records. Data update frequencies fall into three categories: real-time (such as storage location changes, scanned receiving), daily (such as work order picking statistics) and monthly inventory (such as inventory reconciliation). Core fields include material code, lot number, storage location number, inspection status, expiration date and received/dispatched quantity. Common units are pieces, kilograms or cubic meters.
Constraints Imposed on Knowledge Base Retrieval and Recall
Scattered data sources and widely varying update rhythms require knowledge base retrieval to support dataset grouping recall based on data update frequency. Core fields focus on material code, lot number and storage location number. Retrieval must prioritize exact field matching, and avoid using generic keywords to prevent mismatched materials or lots. Diverse daily document types need to be linked to corresponding document sets based on business scenarios, to ensure recalled content aligns with specific audit links such as receiving, inventory or delivery. Core requirements of account-physical consistency and lot traceability mean recall results must include complete inspection status and expiration date information to support quality audit judgments.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches precise fields for warehousing and lot data, avoids generic recall of irrelevant material data |
limit | Top 8–12 entries | Covers the number of material lots to verify in a single audit, while controlling context length |
searchMode | embedding | Prioritizes vector matching based on fields like material code and lot number, to ensure precise recall |
usingReRank | Enabled | Re-ranks recall results by business relevance, prioritizes content related to inspection status and expiration date |
datasetSearchUsingExtensionQuery | Enabled | Adds extended retrieval for fields like lot number and storage location number, covers multi-dimensional query scenarios |
collectionIds | Bind corresponding collections by business scenario | Distinguishes document sets for scenarios such as receiving, inventory, and delivery, avoids cross-scenario recall interference |
Three Common Misconfigurations
- Phenomenon: The knowledge base collection tag entry cannot be found in the interface. Cause: The dataset management permission for the corresponding scenario is not enabled in FastGPT v4.17.0, or no business collection tailored for warehousing audits has been created.
- Phenomenon: The knowledge base associated ID returned in chat does not match the target document. Cause: The
collectionIdsparameter is not filled correctly in the retrieval configuration, causing the recall range to exceed the target collection. - Phenomenon: The HTTP request step in the workflow does not execute, and AI chat is triggered directly. Cause: The correct trigger logic is not set for the workflow node, or the parameter format does not meet the requirements of FastGPT v4.17.0.
How to Verify Correct Configuration
- Enter specific material codes and lot numbers, check whether the recall results include corresponding documents such as receiving slips and inventory counts.
- View the retrieval log, confirm that the
similarityscore of the recall results falls within the 0.75–0.85 range. - Verify the workflow node configuration, confirm that the
collectionIdsparameter is bound to the dataset collection for the corresponding scenario. - Trigger the HTTP request step in the workflow, check the log to confirm that the request has been sent and corresponding business data has been returned.
The values provided here are common starting points, and should be evaluated against the user's own samples.