What the data for this category looks like
The data for this category mainly comes from product digital models and GD&T annotations, process route documents, coordinate measuring machine data, on-site trial feedback, and measurement system records. Daily generated documents include tooling requirement documents, positioning datum schemes, fixture drawings, gauge operation manuals, acceptance reports, calibration records, and change histories. Data updates are performed according to project milestones, product changes, and regular calibrations. Documents contain fields with clear units such as GD&T tolerance values, coordinate measuring machine measurement deviations, and calibration cycles.
What constraints these characteristics impose on the document parsing and chunking link
The document characteristics of this category impose multiple constraints on parsing and chunking. First, for documents containing precise values such as GD&T annotations and coordinate measuring machine deviations, chunking must retain contextual associations to avoid splitting critical parameters and causing information breaks. Second, documents with version associations such as change histories and calibration records must retain logical connections between prior and subsequent content; chunking must not break version sequences. In addition, some documents embed drawings and structured tables, so parsing must support extraction of non-pure text content, otherwise critical tooling positioning information will be lost. Finally, data updated according to projects and changes must ensure that chunked content can accurately associate with document information of corresponding versions, to avoid confusion of tooling parameters from different stages during retrieval.
How to set the configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | This category of documents contains precise parameters such as GD&T tolerances and coordinate measuring machine deviations. Chunking must retain contextual associations to avoid splitting critical numerical values |
chunkSplitter | Split by title + paragraph | Documents have clear section titles such as positioning datums and change histories. Splitting by title preserves section integrity and logical associations |
customPdfParse | Enable structured field extraction | Some documents embed structured content such as GD&T tables and calibration records. Enabling this allows accurate extraction of non-pure text fields |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Large tooling drawing documents have large file sizes. Sufficient time is required for image text recognition and structured conversion during parsing |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Supports uploading large collections of project tooling drawings, meets the demand for batch parsing of multiple documents, and adapts to the upload limit of FastGPT v4.17.0 |
chunkSettingMode | Manual configuration | Chunking rules need to be adjusted for different document types such as acceptance reports and change histories. Manual configuration provides more flexible adaptation to business scenarios |
Three common mistakes
- Symptom: GD&T tolerance fields are empty after PDF parsing, or extracted values do not match the original text. Cause: The
customPdfParseconfiguration is not enabled. The default parsing only extracts pure text and cannot recognize embedded structured tables and tolerance annotations. - Symptom: The interface request returns a parsing timeout status code. Cause: The
PARSE_FILE_TIMEOUT_SECONDSconfiguration value is lower than the actual parsing time. Large tooling drawing parsing is terminated before completion. - Symptom: Chunked content loses section title association information, making it impossible to locate the corresponding positioning datum scheme during retrieval. Cause: The
chunkSplitterconfiguration is set to split by pure characters, not by document title hierarchy, causing section titles to separate from body text.
How to confirm the configuration is correct
- Upload a PDF of a tooling fixture drawing containing GD&T annotations, and check whether the tolerance values and datum information are fully extracted in the parsing results.
- Upload a single collection of tooling documents with a volume exceeding 500 MB, and confirm that the parsing progress does not show abnormal interruptions.
- Randomly select chunked content and verify whether each chunk contains the corresponding section title and coherent body text.
- Initiate a RAG retrieval test, enter a query term containing a positioning datum, and confirm that the returned chunked content is associated with the correct section information.
The values provided here are common starting points and should be measured against the reader's own samples.