This document is written based on the functional logic of FastGPT v4.17.0.
What the data for this category looks like
Production data for precast concrete components comes from BIM or detailed design files, rebar and concrete batch ledgers, mold table and curing kiln operation records, laboratory test reports, and QR code traceability data. Updates are triggered per individual component, per batch of materials, or per curing batch in a kiln.
Daily documents include detailed component drawings, rebar cutting lists, mold inspection sheets, pouring and vibration records, curing logs, and demolding and factory inspection reports. Fields include component ID, rebar specification and model, concrete strength grade, curing duration, demolding strength value, and more. Units include MPa, cubic meters, days, and others.
What constraints do these characteristics impose on document parsing and chunking
The multi-type documents and high-frequency update characteristics of precast concrete components create multiple constraints for document parsing and chunking. First, the range of supported document formats is broad, including CAD detailed drawings, structured cutting lists, and log-style text. Parsing rules must be adapted for different formats. Second, data is closely linked. A single component corresponds to multiple associated documents. Chunking must retain cross-document linked fields such as component ID and batch number to avoid losing contextual association after splitting. Third, there are many industry-specific fields and units, such as concrete strength grade and curing duration. Parsing must accurately preserve field identifiers and units to prevent information distortion. Fourth, batch documents are updated by component and batch. Chunking must align with business aggregation logic to ensure subsequent ledger summarization can integrate data by component dimension.
How to configure the settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the text density of detailed component drawings and rebar cutting lists, retains contextual association within a single document, and avoids splitting business fields into disconnected chunks due to overly short lengths |
chunkSplitter | Trigger chunking rules by document type | Adapts to the diverse document formats of this category: use structured chunking for PDFs exported from CAD, and line-break chunking for log-style text |
customPdfParse | Enabled | Adapts to the extraction of CAD layer information from detailed component drawings, and retains key business content such as component IDs and rebar position annotations in the drawings |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Supports uploading single large detailed drawings or packaged documents of multiple batches, meeting the requirement of uploading batch component documents in a single operation |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Adapts to the parsing time required for large detailed drawings, and avoids task interruption due to parsing timeout |
chunkSettingMode | Manual configuration | Adjust chunking parameters for industry-specific document structures, and avoid losing key business fields when using the general mode |
Three common configuration mistakes
- Phenomenon: After uploading multiple large detailed drawings, the parsing task is interrupted midway, and the parsing progress cannot be restored after restarting Docker. Cause: The
UPLOAD_FILE_MAX_SIZEandPARSE_FILE_TIMEOUT_SECONDSparameters are not adjusted. Large documents exceed the configured limit, or parsing time exceeds the threshold, leading to abnormal task termination. - Phenomenon: Parsed document chunks contain scattered text with no business association, and the model cannot generate an accurate production ledger based on the document content. Cause:
customPdfParseis not enabled, or general chunking rules are used, and the layer association information in detailed component drawings and the table structure of cutting lists are not retained. - Phenomenon: No parsing feedback is provided after uploading a PDF document via the dialog box, and no error logs appear in the backend. Cause:
customPdfParseis not configured to adapt to the PDF format exported from CAD, the parsing engine cannot recognize layer content, causing the task to get stuck, and no error prompt is triggered.
How to confirm the configuration is correct
- Upload a single typical detailed component drawing, and check if the parsed chunks contain key fields such as component ID and rebar specification, with no scattered redundant content.
- Upload a batch of curing log documents, and verify that the parsing task is completed within the preset
PARSE_FILE_TIMEOUT_SECONDS, with no midway interruptions. - Adjust
chunkSettingModeto manual configuration, modify thechunkSizeparameter, and check that the chunk length meets the text density requirements of business documents. - View the backend parsing logs, confirm that the
customPdfParsefunction is enabled, and there are no error messages indicating failed format recognition.
The values provided here are common starting points and should be measured against the reader's own samples.