Reference only; verify against the target FastGPT release and deployment documentation.
What the data for this category looks like
Account data for general contracting of residential construction projects mainly comes from paper forms, electronic submission forms, site acceptance records, material entry account books, and other materials generated during construction. Data updates follow the project progress rhythm: updated daily or weekly during daily construction phases, and updated in batches during phase acceptance phases. Document formats include structured spreadsheets, Word documents with tables, and scanned PDF files. Core fields include project location, material specifications, engineering quantity, corresponding unit price, acceptance personnel, and construction date. Common units include cubic meters, square meters, tons, man-days, and others.
What constraints do these characteristics impose on the "document parsing and chunking" link
The table layouts of structured electronic account books vary. During parsing, cell content and row-column correspondence must be retained, and field combinations spanning multiple rows or columns must not be split. Scanned PDFs account for a notable share of documents, so the OCR parsing process must be enabled; otherwise, valid text cannot be extracted. Core fields are bound to specific units. During chunking, the fields and units of the same account book entry must be kept within the same chunk to avoid semantic breaks. The volume of project documents accumulates as the project progresses. The single chunk length must be adapted to the number of fields in the account book to prevent loss of complete business logic after splitting.
How to set the configuration
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | A single residential construction account entry contains 5-8 fields. This length can fully accommodate a single entry or 2-3 associated account entries, preserving business semantics |
chunkSplitter | Split by paragraph + table | Account documents contain structured tables and paragraph descriptions. Splitting by paragraph preserves table integrity and avoids splitting table rows across chunks |
customPdfParse | Enable OCR mode | Scanned PDF acceptance records exist in projects. Enabling this mode extracts scanned text and adapts to non-editable PDF documents |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Account summary documents for large projects have a large volume. 600 seconds covers the complete parsing process and avoids mid-parsing timeout failures |
chunkSettingMode | Manual configuration | Chunking rules must be adjusted for the structured characteristics of account books. Automatic mode cannot adapt to the special requirements of table and field binding |
UPLOAD_FILE_MAX_SIZE | 1000 MB | A single archived account book document for a residential construction project may contain multiple periods of records. 1000 MB meets the upload requirements for conventional archived documents |
Three common mistakes
- Knowledge base search returns no results, showing no matching content. Cause: Complete account book entries were split across chunks during chunking, making it impossible to match complete business keyword combinations during searches.
- No valid text content is returned after parsing scanned PDFs. Cause: The OCR mode of
customPdfParsewas not enabled, so text in scanned images cannot be recognized. - Parsing task times out and fails, returning status code 504. Cause: The value of
PARSE_FILE_TIMEOUT_SECONDSwas not adjusted. The default timeout duration is insufficient for parsing large account book documents.
How to confirm the configuration is correct
- Upload an account book document containing scanned PDFs, and check whether the parsed text includes the tables and text content from the original document.
- Randomly select a single account book entry, and verify whether all fields and corresponding units of that entry are fully included in the knowledge base chunk content.
- Upload a large archived account book document, wait for parsing to complete, and check whether the task status shows success with no timeout errors.
- Test searching for specific field combination keywords in the account book, and confirm that the returned results include matching complete entries.
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.