What the data for this category looks like
Core data for this category comes from engineering quality and safety supervision systems, on-site construction real-name management platforms, supervision logs, and test reports. Data is updated weekly. Daily processed documents include supervision plans, supervision inspection records, hidden danger rectification notices, quality problem handling forms, and completion acceptance supervision reports. Document structures include fixed-format structured tables, long-text on-site records, and sub-item acceptance reports. Fields cover engineering identifiers, hidden danger levels, rectification requirements, acceptance conclusions, and more. Units vary flexibly based on document type, with no unified mandatory standard.
What constraints these characteristics impose on the document parsing and chunking link
The weekly batch update requirement means the parsing process must support batch task scheduling to avoid overload from single-run processing. The variety of document types means the chunking logic must adapt to mixed scenarios of structured tables and long text, to avoid separating related information between rectification requirements and acceptance conclusions. Documents exported from multiple systems have inconsistent layouts, so custom parsing rules are required to adapt to non-standard PDF formats. Additionally, parsing result accuracy directly affects subsequent hidden danger closed-loop tracking and acceptance management. Keyword fields must not be omitted, and chunked context must remain complete.
How to set configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | This category’s documents include long-text supervision logs and structured tables. Chunks that are too long lose contextual connections. Chunks that are too short break the logical link between rectification requirements and acceptance opinions |
chunkSplitter | Split by paragraph + filter headers and footers | Documents in this category often have fixed quality and safety station name headers and page number footers. Filtering these reduces interference from irrelevant content on chunking |
customPdfParse | Enabled | PDFs in this category are mostly exported from supervision systems, with non-standard layouts and embedded tables. Custom parsing preserves structured fields and image content |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | A single completion acceptance supervision report may exceed 100 pages, so parsing takes a long time. Sufficient processing time must be reserved |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Collections of supervision logs uploaded in batches may have large file sizes, so storage requirements for batch tasks must be accommodated |
Three common misconfigurations
- Symptom: After creating a knowledge base via API, status information for parsing, ready, and failed states cannot be obtained. Cause: The FastGPT knowledge base parsing status query API is not called, or the API callback address is not configured correctly, so batch parsing task progress cannot be tracked.
- Symptom: Set-Cookie fields from documents exported by third-party supervision systems cannot be parsed, so automatic login and document download fail. Cause: Cookie persistence configuration for the custom HTTP request module is not enabled. The code execution module cannot directly call external APIs to obtain session credentials.
- Symptom: After uploading a PDF with on-site construction photos, the parsing result does not include image content or corresponding text descriptions. Cause: The image extraction parameter for
customPdfParseis not enabled, or a model configuration that supports image recognition is not associated.
How to confirm configurations are correct
- Upload a standard hidden danger rectification notice PDF, and check if the parsing result retains structured fields such as rectification time limits and hidden danger levels.
- Call the FastGPT knowledge base parsing status query API, and verify that parsing, ready, and failed status information can be returned correctly.
- Upload a test file containing an HTTP response message with Set-Cookie fields, and confirm that the custom HTTP request module can correctly extract and reuse Cookie fields.
- Upload a completion acceptance supervision report PDF with on-site construction photos, and check if the parsing result includes associated image information or text descriptions.
The values provided are common starting points and should be validated against available sample data.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.