What the data for this category looks like
Typical entities in this category include panel furniture, solid wood furniture, custom home and door enterprises. Daily business documents include order breakdown drawings, cutting lists, edge banding and drilling programs, coating process documents, hardware bill of materials, and trial assembly and packaging records. Data mainly comes from order design systems, sheet and wood batch ledgers, CNC cutting and drilling equipment logs, ERP/MES systems, and moisture content test records. Data is updated via three standard cadences: exclusive documents are generated for each order, production progress is synced daily, and corresponding ledgers are updated for each raw material batch. Document fields include panel size parameters, process parameters, material codes, and test values. Most units are millimeters, cubic meters, and kilograms.
What constraints these characteristics impose on knowledge base retrieval and recall
Custom order-driven multi-variety, small-batch requirements mean the knowledge base must quickly recall exclusive order breakdown and process data per order. Mixed storage of multiple document types requires configuring retrieval rules adapted to different document structures. Frequently updated batch data requires prioritizing recall of latest sheet and equipment logs during retrieval. Retrieval needs for precise parameters such as panel size and hole positions require limiting matching precision of recall results, to avoid production errors from fuzzy matching. Trial assembly records related to on-site installation must be linked to equipment inspection and maintenance scenarios, to ensure recall results cover full-link business.
How to set the configuration
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Adapts to retrieval needs for precise dimensions and process parameters in furniture manufacturing, avoids low-match results interfering with production judgments |
limit | Top 8 entries | Balances recall efficiency and information completeness, covers associated data for multiple panels per single order |
searchMode | embedding | The query examples are natural-language or semantic questions; test this mode with representative data, not as a product default. |
usingReRank | Enabled | Performs secondary sorting on recall results, prioritizes content strongly related to equipment inspection and maintenance scenarios |
datasetSearchUsingExtensionQuery | Enabled | Automatically supplements process and equipment-related extended keywords based on user queries, improves recall coverage |
collectionIds | Bind exclusive collections by business domain | Differentiates datasets such as order breakdown data, equipment logs, and process documents, avoids cross-domain retrieval interference |
Three common errors
- When uploading cutting lists or order breakdown drawings in PDF format to the knowledge base, an "offset out of range" error appears when upload progress reaches around 90%, with a relatively high occurrence rate. The cause is failure to adjust parsing rules for structured PDF documents, leading to block offset calculation exceeding valid ranges during file parsing.
- When calling a workflow via API, a specified knowledge base cannot be referenced, and returns empty fields or permission errors. The cause is incorrect configuration of the
collectionIdsparameter, or failure to bind valid IDs of the corresponding dataset to global variables. - Retrieval results show reference to professional knowledge base content, but the generated answer is completely unrelated to the knowledge base documents. The cause is failure to set a reasonable
similaritythreshold, leading to low-match recall results being included in the context.
How to confirm the configuration is set correctly
- Upload a single structured PDF document, check that the upload process completes normally with no offset-related errors.
- Call a workflow via API, pass the configured dataset ID, check that retrieval results from the corresponding knowledge base can be obtained normally.
- Enter queries related to equipment inspection or maintenance, check that recall results meet the preset threshold rules for matching precision, and that results are strongly related to the query scenario.
- Compare retrieval results with and without the re-ranking function enabled, confirm that sorted results better align with business needs.
The values provided are common starting points, and should be evaluated against the reader's own samples.