What the data for this category looks like
Daily documents for this category include procurement plans, requirement specifications, bidding documents, bid evaluation records, contracts, acceptance documents, supplier complaint materials, and more. They cover two formats: structured tables and unstructured text. Data sources include enterprise procurement platforms, public resource trading platforms, supplier databases, contract management systems, and audit issue databases. The update frequency is daily. Document fields include procurement subject, budget amount, procurement method, scoring method, supplier qualification conditions, performance acceptance standards, and more. Some documents include standardized units.
What constraints these characteristics impose on the "document parsing and chunking" link
Data sources for this category are scattered and updated daily. This requires the document parsing process to support high-frequency batch processing, and avoid slowing down internal office and compliance review workflows. Documents contain both structured tables and unstructured text, so parsing accuracy for both formats must be balanced. Continuous compliance clauses such as scoring methods and supplier qualifications must not be split apart. Errors may lead to risks such as bid rigging and audit rectification. This requires the parsing link to accurately extract key compliance information, and avoid omissions or misjudgments. Core content such as procurement methods and budget basis must be verified during business processes. Parsed chunks must retain the integrity of key information, to reduce the workload of subsequent manual verification.
How to set the configuration
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapt to the paragraph length of documents such as bidding documents and bid evaluation records, and retain the integrity of key compliance information such as procurement methods and scoring methods |
chunkSplitter | Split by paragraph + semantics | Balance the splitting of structured tables and unstructured text in tendering and procurement documents, and avoid splitting continuous compliance clauses |
customPdfParse | Enable table recognition mode | Accurately extract structured table data in bidding documents and bid evaluation records, such as compliance-related fields including procurement budget and scoring items |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Address the parsing time required for large bidding documents and batch bid evaluation records, and avoid interrupting the parsing process due to timeout |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Support importing large-volume documents such as batch supplier databases and historical contracts, and adapt to high-frequency updated data import requirements |
chunkSettingMode | Manual configuration | Adapt to the mixed format characteristics of tendering and procurement documents, and flexibly adjust chunking rules to match the information extraction needs of compliance reviews |
Three common mistakes
- Symptom: After uploading a supplier database Excel file, the data processing result is empty. Cause: The structured table parsing mode is not enabled, or the Excel file encoding format does not meet platform requirements, resulting in the inability to recognize structured fields during parsing.
- Symptom: After batch importing bidding document PDFs, vectorization processing takes too long. Cause: The
chunkSizesetting is too small, resulting in an excessive number of chunks, or batch parsing optimization configuration is not enabled, increasing the computational load of subsequent vectorization. - Symptom: When initiating a question related to procurement compliance, the entered text triggers the document parsing process. Cause: The
trainingTypeconfiguration is incorrect, real-time conversation content is mistakenly included in the training parsing scope, or the trigger logic between knowledge base training and real-time conversation is not correctly distinguished.
How to confirm the configuration is correct
- A standard bidding document PDF is uploaded, and the parsed chunk results are reviewed to confirm that key compliance fields such as scoring methods and procurement budget are not split apart.
- A single supplier database Excel file with a volume not exceeding 1000 MB is uploaded, and it is verified that knowledge base entries can be normally generated after data processing.
- A question related to procurement compliance is initiated, and it is confirmed that the entered text is not included in the document parsing process, only triggering semantic recall.
- 3 or more bid evaluation record documents are batch uploaded, and it is verified that the parsing time meets expectations, with no timeout errors occurring.
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.