What this category of data looks like
Data for this category comes from engineering, production, quality control, and laboratory departments of printed circuit board factories. Daily documents include engineering production instructions, Gerber parsing records, drilling and lamination parameters, chemical solution analysis reports, AOI reports, electrical test reports, and slice analysis reports. Data sources include customer-provided Gerber files, drill tapes, BOM files, CAM files, equipment parameters, bath solution test results, AOI data, electrical test data, and laboratory test data. Update frequency follows per production batch, per shift, or continuous collection. Documents contain numeric parameters such as aperture, temperature, and concentration, as well as structured test records and unstructured engineering descriptions.
What constraints these characteristics impose on the document parsing and chunking step
Multi-source, heterogeneous document formats and high-frequency update cycles require parsing systems to support multiple file types including Gerber, BOM, Word, and PDF. Systems must also adapt to small-batch, high-frequency parsing demands. A large number of associated technical parameters in documents — such as the binding between drilling aperture and lamination sequence — require retaining context associations during chunking. This prevents loss of critical technical logic after splitting. Some documents like AOI reports and slice reports include image annotations. Parsing must simultaneously extract the binding relationship between images and their corresponding text descriptions. This ensures complete information after chunking. Document structures vary across different production batches. Chunking rules must adapt to non-fixed-format engineering documents. This avoids forced splitting that damages the integrity of technical expressions.
How to configure the settings
The following configurations are based on FastGPT v4.17.0:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the association length between parameters and context in PCB documents, avoids splitting critical technical paragraphs |
chunkSplitter | By paragraph + semantic boundary | Adapts to non-fixed-format documents such as Gerber parsing records and engineering instructions, retains associations between parameters and descriptions |
customPdfParse | Enable image-text binding parsing | Adapts to image annotations in AOI reports and slice reports, extracts text descriptions associated with images |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Adapts to the parsing duration of batch Gerber files and large BOM documents, avoids timeout failures |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Adapts to the demand for batch uploads of packaged customer data, avoids interruptions during large file parsing |
chunkSettingMode | Manual configuration | Adjust chunking rules for different types of bid documents such as quotation sheets and engineering instructions, adapts to specific use cases |
Three common misconfigurations
- Symptom: Image-containing Word or PDF documents are uploaded to the knowledge base, and associated annotation content cannot be retrieved during question-and-answer testing. Cause: The
customPdfParseconfiguration is not enabled. Only plain text is extracted during chunking, and text descriptions bound to images are not retained. - Symptom: After batch uploading packaged Gerber files, the parsing process shows a timeout status code. Cause: The
PARSE_FILE_TIMEOUT_SECONDSsetting value is too small, and sufficient parsing duration for large packaged multi-file packages is not allocated. - Symptom: Parsed chunked content is missing critical numeric parameters such as drilling aperture and lamination temperature. Cause: The
chunkSizesetting value is too small, which cuts off the association between parameters and contextual technical descriptions during splitting.
How to confirm the configuration is correct
- Upload a single AOI report document with image annotations, view the parsed chunked content, and confirm that the image’s text annotations and corresponding paragraph appear in the same chunk.
- Upload batch packaged Gerber and BOM files, verify that the parsing process has no timeout errors, and confirm that all files have completed parsing.
- Adjust the chunking rules for different types of bid documents, and check whether the chunked results retain the association between parameters and their context.
- Submit test questions related to technical parameters, and confirm that the system can retrieve the content of the corresponding chunks.
The values provided here are common starting points, and should be evaluated against the reader's own samples.