What the data for this category looks like
This category’s data primarily comes from order design systems, sheet and wood batch records, CNC cutting and drilling equipment logs, ERP/MES systems, and moisture content testing equipment. Daily documents include order disassembly drawings, cutting lists, edge banding and drilling programs, coating process documents, hardware accessory lists, trial assembly and packaging records. Data update frequency is adjusted based on orders, daily cycles, or material batches. Fields include panel dimensions (millimeters), hole position parameters, coating temperatures, hardware models, sheet moisture content, and more. Some data is structured in tables or CNC program code.
Constraints for Knowledge Base Retrieval and Recall From These Category Characteristics
Custom order-driven production models result in a large number of highly personalized panels. Retrieval must accurately match key parameters such as panel ID and dimensions to avoid retrieving irrelevant general process documents. Multi-source, high-frequency update data requires the knowledge base to support incremental synchronization to ensure retrieval results align with current production batches. Document types include structured tables and CNC code, so retrieval must balance keyword matching and semantic understanding, while also controlling the context length of retrieved results to avoid exceeding model processing limits. Errors directly affect production installation and product quality, so retrieval accuracy requirements are high, and content strongly related to the current order must be prioritized for retrieval.
How to Configure the Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Requires precise matching of structured parameters such as panel dimensions and hole positions to avoid retrieving irrelevant coating or hardware documents for the current order |
limit | Top 10 results | Custom orders have a large number of panels, so multiple dimensions of process requirements must be covered, while controlling the total context length to not exceed model processing limits |
searchMode | mixedRecall | Furniture manufacturing documents include both structured process parameters and non-standard on-site operation descriptions; hybrid search improves retrieval coverage |
usingReRank | Enabled | After retrieving multiple documents, sort them by order batch and sheet type, prioritize matching core requirements of current production tasks, and reduce risks of color difference and dimension errors |
datasetSearchUsingExtensionQuery | Enabled | Automatically supplement context keywords such as order batch and sheet moisture content to strengthen the relevance of retrieval to current production scenarios |
collectionIds | Configured by production workshop/order type | Process documents for different workshops are independent, preventing retrieval of irrelevant cutting programs or coating processes across workshops |
Three Common Misconfigurations
- Issue: Retrieval results include workshop process documents unrelated to the current order, and the number of results exceeds expectations. Cause:
collectionIdsis configured as the full knowledge base collection, without grouping by production scenario. - Issue: The interface displays format errors when uploading a cutting list, and import cannot be completed. Cause: The document does not include required category fields such as panel ID and dimensions, or does not use a supported structured format.
- Issue: Empty results are returned when calling knowledge base retrieval, and the knowledge base bound to global variables does not take effect. Cause: Incorrect variable format was used when dynamically assigning
collectionIds, and the knowledge base ID associated with the current order was not matched.
How to Verify Proper Configuration
- A test custom panel cutting list is uploaded. Hole position parameters for a specific panel are retrieved, and whether the corresponding document appears in retrieval results is verified.
- Retrieval logs are viewed to confirm that the call status of
usingReRankanddatasetSearchUsingExtensionQueryis enabled. collectionIdsare dynamically bound to the specified workshop’s collection ID. Edge banding process documents for that workshop are retrieved, and only content within the corresponding collection is confirmed to be returned.- The
similarityparameter value range is adjusted, and whether the relevance of retrieval results meets current production accuracy requirements is validated.
This document was developed for FastGPT v4.17.0.
The configuration values listed above are common starting points and should be validated against specific samples for the deployment environment.