What this category of data looks like
Daily documents for this subject typically include mold design task sheets, DFM reports, part drawings, machining process cards, trial mold reports, mold modification notices, and mold history records. Data comes from customer product models, mold flow or forming simulation data, CAD/CAM files, machining and measurement data, and trial mold samples. Updates are carried out per project, with synchronized updates after each trial mold or mold modification. Documents include structured fields and unstructured text, and some files contain quantitative parameters such as dimensions and working hours.
What constraints do these characteristics impose on the knowledge base retrieval and recall link
Documents for this category include structured process cards, unstructured trial mold reports, and CAD drawings, and are updated on demand per project. Mixed-format data requires retrieval to adapt to the recall logic of different document types, to avoid excessive truncation of drawing text. The per-project update feature requires the retrieval link to support incremental synchronization, without the need for a full knowledge base rebuild each time. In addition, core business links rely on time-series data such as trial mold defects and mold modification records, so retrieval needs to prioritize recall of high-relevance project-level document fragments.
How to configure
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the mixed structured and unstructured characteristics of mold-related documents, balancing recall precision and coverage |
limit | Top 8 entries | Matches the volume of documents under a single project, avoiding context overload that impacts retrieval accuracy |
searchMode | mixedRecall | Combines keyword and semantic retrieval, adapting to documents of different formats such as part drawings and process cards |
usingReRank | Enabled | Re-ranks recall results to prioritize returning highly relevant fragments related to trial mold defects and mold modification requirements |
datasetSearchUsingExtensionQuery | Enabled | Expands retrieval terms based on the original question, covering synonymous expressions for mold design terminology and trial mold defects |
collectionIds | Project-bound collection IDs | Isolates mold documents from different projects, avoiding cross-project recall interference |
Three common mistakes
- Issue: Format error occurs when uploading knowledge base files. Cause: Failed to distinguish upload rules for structured process cards, unstructured trial mold reports, and CAD drawings, with mixed uploads without format adaptation.
- Issue: Variable references in the
[{datasetId: xxx}]format do not take effect. Cause: Did not follow the FastGPT v4.17.0 variable reference specification, or did not bind the access permission for the corresponding dataset in the call chain. - Issue: The knowledge base continues to show the indexing status after upload. Cause: The total volume of mold documents uploaded in a single batch is too large, and uploads were not split by project, causing the indexing process to time out.
How to confirm the configuration is complete
- Upload a single mold design task sheet and trial mold report, and check whether the indexing progress completes within a reasonable time frame.
- Enter the question "How to address shrinkage defects during trial mold" and verify that the retrieval results recall relevant fragments from the corresponding trial mold report and mold modification notice.
- Initiate a call after configuring the variable reference, and check whether the returned results include document content from the bound dataset.
- Upload a mold part drawing to the knowledge base, initiate a relevant question retrieval, and confirm that the corresponding image is displayed in the results.
The values provided here are common starting points and should be measured against the reader's own samples.