What this category's data looks like
The core data for this category comes from customer drawings, special characteristic lists, ERP/MES systems, measurement and inspection tool data, equipment process parameters, and supplier quality certificates. Daily generated documents include control plans, process flowcharts, work instructions, first-piece and patrol inspection records, SPC reports, PPAP materials, 8D reports, and others. Some documents have structured tables and fixed fields such as special characteristic numbers, process tolerances, inspection frequencies, with units covering industrial standard units including millimeters, newtons, hours. Data update frequency is per batch, per shift, or continuous collection. A single batch file package may contain multiple related documents.
Constraints on document parsing and chunking
Multi-source and heterogeneous document types require parsing tools to adapt to different formats, such as SPC reports with tables and work instructions combining text and images. High-frequency updates and batch upload requirements mean the parsing process must support large-file batch processing and fast response. Fixed fields and industrial unit standards require that chunking retains field relevance, and avoids splitting cross-page logical units. Additionally, constrained by vehicle customer quality systems, parsing results must accurately restore special characteristic information from documents without omitting key parameters.
Configuration Settings
These configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | The documents contain short fields such as tolerance values and inspection frequencies. Excessively large chunks will mix in unrelated content, while excessively small chunks will destroy logical coherence |
chunkSplitter | Semantic chunking | Control plans and 8D reports have clear logical hierarchies. Semantic chunking can retain chapter integrity and adapt to business audit logic |
customPdfParse | Enable structured parsing | PPAP materials and SPC reports are mostly PDF documents with tables. Structured parsing can accurately extract special characteristics and parameter fields within tables |
UPLOAD_FILE_MAX_SIZE | 1000 MB | A single batch upload PPAP package may contain multiple drawings and reports, so the total size of the single batch file collection must be accommodated |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Large multi-page SPC reports or complete PPAP materials take a long time to parse, avoiding parsing process interruption due to timeout |
chunkSettingMode | Manual configuration | It is necessary to adjust chunking parameters for different document types such as work instructions and 8D reports to adapt to the field structure of each document |
Three Common Configuration Mistakes
- Symptom: No special characteristic fields appear in parsing results after uploading PPAP materials. Cause:
customPdfParseis not enabled, and the PDF structured parsing function is not activated, so structured fields within tables cannot be extracted. - Symptom: Parsing failure is returned after calling the workflow API to upload images. Cause: The correct format of image parameters is not passed as required by the documentation, and associated settings for image parsing are not configured, resulting in the inability to recognize images.
- Symptom: The root cause analysis section of an 8D report is split into multiple chunks. Cause: Fixed character chunking mode is used, and switching to semantic chunking is not performed, forcibly splitting logically coherent chapters by character length.
How to Verify Correct Configuration
- Upload a standard PPAP material, review the parsed chunk content, and confirm that core fields such as special characteristic numbers and tolerance values are fully retained.
- Call the file upload API, upload a batch file package within 1000 MB, confirm that no 413 status code is returned, and the parsing task starts normally.
- Switch to semantic chunking mode, upload a complete 8D report, and confirm that logical chapters such as root cause analysis and corrective measures are not forcibly split.
- Check the parsing task log, confirm that the parsing time of a single file does not exceed 600 seconds, and no timeout error prompts are present.
The values provided are common starting points and should be evaluated against samples relevant to the specific deployment.