What the data for this category looks like
Core data for furniture and wood products manufacturing comes from order design systems, sheet and lumber batch ledgers, CNC cutting and drilling equipment logs, ERP/MES systems, and moisture content test records. Daily output documents include custom cut plans, cutting lists, edge banding and drilling programs, coating process specifications, hardware accessory lists, and trial assembly and packaging records. Data updates run per order, daily, or per material batch. Each document has unique fields such as panel size, texture color code, hole position parameters, and hardware model. There is no fixed field structure for a universal standard BOM.
Constraints on Knowledge Base Retrieval and Recall
Custom orders drive high panel counts with scattered size parameters. Retrieval must match multi-dimensional panel attributes. High-frequency updates per order, daily, or per material batch require the retrieval pipeline to support real-time sync of latest sheet batches and process parameters. Unstructured documents like cut plans and CNC programs coexist with structured documents like hardware lists and cutting plans. This requires support for hybrid retrieval modes. Custom scenarios have no universal fixed BOM. This means generic retrieval logic must avoid redundant recall, and focus on the current order’s exclusive parameters and batch information.
Configuration Settings
These configurations are adapted for FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Matches high-precision parameters such as panel size and hole positions, and avoids recalling irrelevant panel data |
limit | Top 8 entries | The number of panels per single order typically does not exceed 50. Limiting recall entries avoids context overload |
searchMode | mixedRecall | Adapts to hybrid retrieval requirements for structured parameters and unstructured documents |
usingReRank | Enabled | Reorders recall results by order relevance, optimizing retrieval accuracy in custom scenarios |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands search terms to cover industry-specific terminology such as cut list terms and coating processes |
collectionIds | Knowledge base collection bound to the current order | Only recalls process, sheet, and hardware data associated with the current order, avoiding cross-order interference |
Three Common Mistakes
- Phenomenon: Retrieval returns large volumes of cross-order panel data, and cannot accurately match the exclusive parameters of the current custom order. Cause: The retrieval knowledge base collection is not limited via
collectionIds, so the recall scope covers all historical data. - Phenomenon: When calling the FastGPT chat API, it is not possible to specify only retrieval of the knowledge base associated with the current order. Cause: The correct knowledge base collection parameter is not configured in the API request, so the retrieval scope is not limited.
- Phenomenon: Knowledge base import tasks fail when importing web-based coating process documents that require login permissions. Cause: No login verification skip logic is configured for web crawling, so document content cannot be obtained normally.
How to Verify the Configuration
- Import the current order’s cut plan and cutting list, run a retrieval, and check if the returned panel parameters match the current order.
- Call the chat API with parameters for the specified knowledge base, and confirm that the returned results only include document content from that knowledge base.
- Import a web document that requires login permissions, and check if the knowledge base import task execution status is normal.
- After enabling the re-ranking function, check if the recall results are sorted to prioritize matching the current order’s exclusive process and material parameters.
The values provided are common starting points, and should be evaluated against the reader's own samples.