What the Data for This Category Looks Like
Core data for this category originates from engineering orders, sand, gravel, cement admixture testing, moisture content measurements, mixing plant control systems, and project laboratories. Daily documents include mix proportion notices, opening identifications, production and transport slips, raw material inspection records, test block ledgers, and strength reports. Data updates follow three schedules: per vehicle production slip, per batch of raw material tests, and daily summary reports. Document fields include batching dosage, slump index, test block strength value, inspection batch number, pouring location, and additional relevant fields. Common units are kilograms per cubic meter, megapascals, millimeters, and similar. There is no unified fixed document format; some files are a mix of scanned documents and spreadsheets.
Constraints Imposed on Document Parsing and Chunking
Frequently updated single-batch documents require chunking aligned to business granularity, to prevent mixing content across batches. Mixed-format documents (spreadsheets, scanned PDFs) demand parsing tools that support multiple file types, while preserving original fields and unit information. Batch-updated test block strength records and mix proportion data require retaining contextual links during chunking, to ensure complete single-batch correction information is retrievable during searches. Bulk-uploaded monthly summary reports require parsing timeout settings adjusted for large document volumes, and upload file size limits sized to meet bulk document storage needs.
How to Configure Settings
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Matches business granularity of single-batch production slips and test block records, avoids cross-batch content mixing, and ensures complete retrieval context |
chunkSplitter | Split by paragraph + table | Adapts to mixed-format documents including mix proportion notices and raw material inspection reports, preserves table fields and unit information |
customPdfParse | Enable OCR mode | Processes scanned opening identification documents and test block records, prevents incorrect extraction of scanned text |
PARSE_FILE_TIMEOUT_SECONDS | 120–180 seconds | Adapts to parsing time for bulk uploads of multiple vehicle production slips, prevents single-run parsing timeout failures |
UPLOAD_FILE_MAX_SIZE | 500 MB | Supports monthly summary inspection report packages, meets bulk document upload requirements |
chunkSettingMode | Manual configuration | Adjusts chunking parameters separately for single-batch documents and summary reports, adapts to frequently updated business data |
Three Common Misconfigurations
- Uploading a scanned raw material inspection report results in empty displayed fields. The cause is that OCR mode for
customPdfParseis not enabled, leading to scanned text failing to be correctly identified and extracted. - Uploading multiple vehicle production and transport slips results in parsing timeout. The cause is that
PARSE_FILE_TIMEOUT_SECONDSis set too low, failing to adapt to parsing time for bulk documents. Restarting the server cannot resolve underlying configuration issues. - Data is missing or incomplete after importing or exporting the knowledge base. The cause is that chunking rules are not configured per business batches, leading to cross-batch content being merged during chunking. Subsequent searches lose single-batch mix proportion correction information.
How to Verify the Configuration Is Correct
- Upload a single scanned raw material inspection report, check if parsed content includes original fields such as inspection batch number and aggregate indicators, to confirm the
customPdfParseconfiguration takes effect. - Upload 5 consecutive production and transport slips, check if parsing time falls within the range set for
PARSE_FILE_TIMEOUT_SECONDS, with no timeout error prompts. - Export the configured knowledge base chunked data, check if each chunk contains complete single-batch mix proportion or test block records, with no cross-batch mixing.
- Upload a monthly report package at the maximum allowed size, confirm upload proceeds normally with no file size limit error prompts.
The values provided here are common starting points, and should be evaluated against the reader's own samples.