Reference only; verify against the target FastGPT release and deployment documentation.
What this category of data looks like
Documents related to urban water utilities originate from daily operations and compliance workflows, including pipeline inspection logs, water quality test reports, drainage permit application materials, pump station operation logs, and more. Data sources include water utility operation systems, scanned on-site paper records, and exported files from compliance declaration platforms. Update frequency varies by document type: inspection logs are updated daily, water quality test reports are updated weekly or monthly, and compliance declaration materials are submitted as needed. Document structures contain fixed structured fields such as test points, indicator names, values, and units, alongside free-text operation notes. Some documents include attached on-site photos and video files.
Constraints on document parsing and chunking
For fixed structured fields, retain the association between fields and their corresponding content during parsing and chunking to avoid splitting the value and unit of the same test data. The mixed structure of short entries and long text requires chunking tools to balance the integrity of short structured data and the semantic coherence of long operation notes. Documents with attached image and video attachments require the parsing process to associate attachments with corresponding content in the main document. High-frequency updated inspection documents require chunking configurations to adapt to batch parsing efficiency and time limits.
How to configure the settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the mixed structure of water utility documents containing both short test data entries and long operation notes, avoiding splitting semantic units |
chunkSplitter | By paragraph + semantic boundary | Adapts to scenarios with mixed structured reports and free text, retaining the integrity of fields and paragraphs |
customPdfParse | Enabled | Adapts to scenarios where water utility documents often contain mixed fixed-format PDF reports and scanned copies, improving the accuracy of structured field extraction |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Addresses parsing time consumption for long-text operation logs and multi-page batch documents, avoiding mid-process timeout interruptions |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Supports batch upload of multiple water utility operation documents and audio-video attachments, adapting to project-level document aggregation scenarios |
chunkSettingMode | Semi-automatic mode | Allows separate adjustment of chunking parameters for different types of water utility documents, such as test reports and inspection logs, adapting to diverse document structures |
Three common misconfigurations
- Phenomenon: After uploading surveillance video files from water utility sites, parsing results are empty or return timeout errors. Cause: Parsing parameters suitable for audio-video content are not configured, and the value of
PARSE_FILE_TIMEOUT_SECONDSis insufficient to complete the long video parsing process. - Phenomenon: When calling the workflow API to upload a water quality test report in image format, a parameter error prompt is returned. Cause: The corresponding fields for the image file are not passed according to the interface specification, or
customPdfParseis not enabled to adapt to structured text embedded in images. - Phenomenon: When searching water utility documents, consecutive content of the same inspection record is split into multiple unrelated results. Cause:
chunkSizeis set too small, splitting consecutive operation notes and test data of the same point, destroying semantic integrity.
How to verify correct configuration
- A typical monthly water quality test report may be uploaded, and the parsed text checked to confirm complete retention of structured fields such as test points, indicator names, values, and units.
- The batch parsing interface may be called to upload 3 to 5 different types of water utility documents, and checked for timeout errors, with the chunk length of parsing results confirmed to match the preset
chunkSizerange. - Chunk results in the vector store may be reviewed to confirm that operation notes of the same paragraph are not split, and that the value and unit of a single test data are not split across chunks.
- A small surveillance video file may be uploaded for testing, to confirm that the parsing process triggers normally and no parameter errors are returned.
The values provided are common starting points and should be evaluated using samples appropriate for the specific deployment.
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.