What the data for this category looks like
Core data for ready-mixed concrete production comes from commercial concrete batching plant control systems, project laboratories, engineering orders, and raw material inspection records. Daily documents include mix proportion notices, opening identifications, production transport slips, raw material inspection reports, test block records, and strength reports. Fields cover sand and gravel cement ratios, moisture content, slump, strength grades, and more. Units include kg/m³, %, mm, MPa, and others. Data is updated per vehicle production batch, each batch of incoming raw materials, and daily laboratory test archiving.
What constraints do these characteristics impose on the document parsing and chunking link
Production transport slips updated per vehicle require chunking logic that can quickly handle high-frequency new documents, and avoid excessive chunk size that causes parsing delays. A large number of structured tables in documents, such as mix proportion tables and test block records, require accurate extraction of cell content to prevent chunking from truncating table rows or columns. 28-day age data from test block strength reports needs to retain associated test dates and batch information, to avoid chunking splitting critical context. Multiple batches of raw material inspection data must be bound by batch, ensuring that traceability can link to corresponding raw material incoming records.
How to configure
Configuration items under FastGPT v4.17.0 are listed below:
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Single-batch documents for ready-mixed concrete include complete parameters for batching and testing. This range avoids truncating key fields while ensuring coherent context within each chunk |
chunkSplitter | Paragraph + table recognition mode | Documents contain unstructured test records and structured tables. Chunking by paragraph preserves context, and the table recognition mode can fully extract batch information within cells |
customPdfParse | Enable pdf-marker parsing | PDF documents for ready-mixed concrete are mostly test reports with nested tables. This parsing mode can accurately restore table structures and avoid content loss |
PARSE_FILE_TIMEOUT_SECONDS | 120–180 seconds | Complete single-batch archives include multi-page reports and batching slips, which require longer parsing time. This range prevents task timeouts |
chunkSettingMode | Manual configuration mode | Chunking parameters need to be adjusted for different document types such as mix proportion slips and test block reports, to adapt to differences in document structures across this category |
Three common mistakes
- Phenomenon: After uploading a CSV-format raw material inspection table, search results only return more than 30 data entries, omitting remaining batch entries. Cause: Table recognition mode is not enabled. The default chunking logic truncates subsequent table rows and fails to fully extract inspection parameters for all batches.
- Phenomenon: When calling the custom PDF parsing interface, a 404 error is returned, and the
/v2/parse/filepath cannot be found. Cause: The v2 version of the pdf-marker service is not deployed, and the service mapping corresponding tocustomPdfParseis not enabled in the platform configuration. - Phenomenon: After batch uploading multiple production transport slips, the chunking results do not associate corresponding documents by batch. Cause: Batch information is not bound using the correct multi-file parameter format, resulting in loss of the association between chunks and original documents.
How to confirm the configuration is correct
- Upload a complete mix proportion notice PDF, check whether the parsed chunks fully retain continuous key parameters such as sand and gravel ratios, moisture content, and slump, with no table splitting or content truncation.
- View the running logs of the parsing task, confirm that the configuration of
PARSE_FILE_TIMEOUT_SECONDSdoes not trigger a timeout error, and the final status of the task is successfully completed. - Pass in 3 test block strength reports from different batches, check whether the returned chunking results are distinguished by batch, and each chunk includes the test date and strength data of the corresponding batch.
- Adjust
chunkSizeto 1000 characters, upload a daily production summary document, and confirm that the chunking does not truncate continuous paragraphs containing all production batches of the day.
The values provided are common starting points and should be evaluated using the samples relevant to the specific deployment.