What this category of data looks like
Data for municipal water utility application materials and case review comes primarily from water utility operation end application systems, pipe network inspection archived files, and government approval receipts. Data updates trigger with business processes. Monthly water supply permit applications update once per month. Daily inspection cases update in real time when submitted on-site. Documents include fixed-format PDF application forms with fillable fields and stamped pages, mixed-text-and-image inspection case PDFs, and Excel ledgers with pipe network numbers, pressure values, and inspection durations. Fields use water utility-specific units such as MPa, m³, and pipe diameter in mm. Some documents include on-site photos of pipes and valve statuses.
What constraints these characteristics impose on the document parsing and chunking step
Fixed-format application forms have unified headers and cross-page fillable fields. Chunking must avoid splitting cross-page fill logic, otherwise field associations will break. Mixed-text-and-image inspection cases include both text descriptions and on-site photos. If parsing does not distinguish text and image blocks, text and corresponding photos will separate. Excel ledgers have numeric fields with specific units. Chunking must retain the binding between fields and units, otherwise subsequent searches cannot match the corresponding water utility parameters. Real-time updated case files may have large single submission volumes, which increases overall parsing time.
How to set the configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Single-section fill content and inspection descriptions in municipal water utility application materials are moderate in length. This avoids splitting fields that belong to the same business logic |
chunkSplitter | Split by title and paragraph boundaries | Water utility documents have fixed headers and inspection paragraphs. Splitting by boundaries preserves business logic integrity |
customPdfParse | Enable image-text separation mode | Inspection cases include on-site photos and text descriptions. Separation allows corresponding content to be associated separately |
UPLOAD_FILE_MAX_SIZE | 500 MB | Monthly application summary files submitted in a single batch may include multiple cases. This reserves sufficient upload space |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Large summary files take longer to parse. This prevents mid-parsing timeout interruptions |
chunkSettingMode | Manual configuration | Water utility documents have fixed formats and specific units. Manual configuration adapts to field binding requirements |
Three common mistakes
- When uploading an Excel file, the parsed result has empty image fields. The cause is that the image extraction switch in custom parsing is not enabled. Only text content is extracted.
- After parsing a PDF, text and on-site photos are in separate chunks. They cannot be associated with the corresponding inspection description. The cause is that the image-text separation configuration for
customPdfParseis not enabled. - Parsing a file larger than the set threshold triggers a timeout error with status code 504. The cause is that the
PARSE_FILE_TIMEOUT_SECONDSparameter is not adjusted based on file size. Parsing time exceeds the preset value.
How to confirm the configuration is correct
- Upload a typical monthly application PDF. Check if headers and fillable fields in the parsed result are fully extracted.
- Upload an inspection case that includes on-site photos. Confirm that the association between text and photos is correct.
- Upload a large summary file under 500 MB. Verify that parsing completes within a reasonable timeframe without errors.
- Split an Excel ledger file. Confirm that bound fields such as pipe network numbers and pressure values are not split into different chunks.
The values provided here are common starting points. They should be evaluated against the reader's own samples.
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.