Reference only; verify against the target FastGPT release and deployment documentation.
What the data for this category looks like
The ledger data for communication towers and infrastructure mainly comes from Excel files exported from operation and maintenance inspection systems, scanned PDFs taken during on-site inspections, PDFs of infrastructure design drawings, and Word documents for rectification notices. Daily inspection reports are updated weekly, basic ledgers are synchronized quarterly, and rectification reports are updated when events are triggered. Document structures include structured tables (such as tower number, longitude and latitude, tower height, inspection date), unstructured inspection descriptions, and PDFs with embedded scanned images. Fields and units include tower height (meters), longitude and latitude (decimal degrees), installation year (YYYY), inspection result (qualified/unqualified), etc. Some overseas base station ledgers include multilingual annotation content.
Constraints on the Document Parsing and Chunking Link
Scanned PDF documents with embedded images cannot extract text directly. OCR recognition is required to obtain valid content, so the parsing link must support custom OCR mode. Structured Excel ledgers contain multiple associated fields. Chunking must retain field associations for the same row to avoid inability to perform associated retrieval after splitting. Large infrastructure design drawing PDFs have large individual file sizes, leading to longer parsing times, which imposes requirements on upload size limits and parsing timeout settings. Inspection reports are updated weekly, so batch upload scenarios need to adapt to high-frequency parsing requirements to avoid task failure due to timeout. Some overseas base station ledgers include multilingual text, so parsing must adapt to multilingual recognition logic.
How to Set Configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
customPdfParse | Enable OCR mode | Most communication tower inspection reports are scanned PDFs embedded with on-site photos, so OCR is required to extract text content from images |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Inspection PDFs with more than 50 pages or large design drawings take a long time to parse; this range covers parsing needs for most scenarios |
chunkSize | 800–1200 characters | Ledger data contains multiple associated fields; this range balances context completeness and retrieval accuracy, avoiding splitting associated data across fields |
chunkSplitter | Paragraph + fixed character boundary | Balances row data from structured Excel and unstructured inspection descriptions, ensuring chunking logic adapts to different document types |
UPLOAD_FILE_MAX_SIZE | 500 MB | Individual large infrastructure design drawing PDFs can reach 200–400 MB; this setting covers upload needs for most large documents |
chunkSettingMode | Manual configuration | Chunking parameters need to be adjusted for different document types such as inspection reports, ledger Excel files, and design drawings; manual configuration enables fine-grained control |
Three Common Configuration Mistakes
- Parsing results are empty or only contain garbled characters after uploading scanned inspection PDFs: OCR mode for
customPdfParseis not enabled, only image binary data is parsed without extracting valid text. - Fewer than expected data entries are returned when searching Excel ledgers:
chunkSizeis set too small, splitting associated fields in the same row, orchunkSplittersplits by row, causing retrieved content across fields to be split and lost. - Timeout errors are triggered when parsing large PDFs:
PARSE_FILE_TIMEOUT_SECONDSis set lower than the actual parsing time, and the default configuration cannot cover the parsing duration of multi-page large documents.
How to Verify Configuration Correctness
- Upload a single inspection PDF embedded with scanned images, check if the parsed text content includes on-site inspection text descriptions and detection data, confirming that the OCR mode of
customPdfParseis active. - Upload an Excel ledger containing multiple associated fields, randomly select a record that includes tower number, inspection date, and inspection result, confirm that all associated fields are included in the returned results after searching, verifying that the chunking configuration is reasonable.
- Upload a single design drawing PDF with a volume exceeding 200 MB, confirm that the upload is successful and the parsing task completes normally, verifying that the
UPLOAD_FILE_MAX_SIZEconfiguration is sufficient. - Upload a 50-page inspection PDF, check the parsing task's time consumption log, confirm that the time does not exceed the set value of
PARSE_FILE_TIMEOUT_SECONDS, verifying that the timeout configuration is reasonable.
The values provided are common starting points and should be measured 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.