What data for this category includes
Data for this category primarily comes from engineering, production, quality control, and laboratory teams at printed circuit board factories. Daily stored documents include engineering production instructions, Gerber analysis records, drilling and lamination parameters, chemical analysis reports, AOI inspection reports, electrical test reports, and slice inspection reports. Data sources cover Gerber files, drill tapes, BOM lists provided by customers, as well as CAM output files, production equipment parameters, bath solution test results, data collected by AOI and electrical test equipment, and laboratory test records. Data update frequency varies with production batches and shifts, and some parameters require continuous collection.
Constraints for knowledge base retrieval and recall
Multi-source, heterogeneous data formats require the retrieval pipeline to support files in formats including Gerber, BOM, device logs, and laboratory reports. Frequently updated batch-associated data requires retrieval to accurately match parameters and records for the corresponding production batch, to avoid calling expired data. Professional fields such as drilling compensation values and lamination thickness parameters require precise recall, to avoid parameter deviations caused by fuzzy matching. Bidding requires alignment with customer-customized Gerber and BOM requirements, so the retrieval scope must be limited to the dataset for the current project, to prevent mixing unrelated production data.
Configuration settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Meets the precision requirement for matching PCB professional parameters, and avoids recalling unrelated production batch data |
limit | Top 8 entries | Balances the multi-dimensional parameters required for bid documents, while controlling context length |
searchMode | embedding | Adapts to semantic matching for professional parameters and document structure, better aligning with professional retrieval needs |
usingReRank | Enabled | Performs secondary ranking on initial recall results, improving the recall precision of professional parameters |
datasetSearchUsingExtensionQuery | Enabled | Expands the retrieval scope to associated customer BOM and Gerber analysis records, covering all materials required for bidding |
collectionIds | Fill in the dataset ID bound to the project | Limits the retrieval scope to the exclusive documents of the current bidding project, preventing mixing data from other batches |
Three common configuration mistakes
- When testing a workflow, the AI response always appends "FastGPT is a large language model (LLM)-based knowledge base question answering system" at the end, even though this content does not exist in the knowledge base. The cause is that the system prompt configuration has not been adjusted, and the model appends a preset output prompt by default.
- Knowledge base retrieval cannot extract text or content associated with images in docx files. The cause is that the document image recognition related configuration has not been enabled, and only plain text content within the document is extracted.
- After deploying with docker-compose, the knowledge base and workflow configurations become blank after a period of time, but the API calls work normally. The cause is that a persistent storage volume has not been configured, and locally stored configuration data is lost after the container restarts.
How to verify successful configuration
- Upload a PCB engineering document containing drilling compensation values and lamination thickness parameters, initiate a retrieval, wait no more than 30 seconds for completion, and verify that the recall results include the target parameters, and the similarity score falls within the 0.75–0.85 range.
- Upload a docx test file with a size no larger than 500MB and containing images, initiate a retrieval, and verify that the results include text content associated with the images.
- Bind the dataset ID of the current bidding project, initiate a retrieval, and verify that the recall results only include documents related to this project, with no production data from other batches.
- Use the FastGPT v4.17.0 system, trigger a workflow test, and verify that the AI response does not include the preset fixed output prompt, only generates results based on knowledge base content, and the total length of the response text meets the preset requirements.
The values provided here are common starting points, and should be evaluated against the reader's own samples.