What the data for this category looks like
Typically, business data for IT procurement and supplier management comes from procurement systems, supplier quotation documents, budget spreadsheets, and contract systems. Updates are triggered per individual procurement process, and monthly procurement data is also aggregated on a monthly basis. Commonly processed documents include purchase requisitions, comparison records, formal procurement contracts, and acceptance documents. These documents contain fields such as purchase item numbers, budget amounts, supplier qualifications, and acceptance criteria, with common units including ten thousand yuan, units, service cycles, and similar.
Constraints on the document parsing and chunking workflow
Documents in this category are mostly structured or semi-structured business forms, with fields tightly bound to business logic. When splitting content, the association between fields must be retained to avoid damaging the integrity of key information such as purchase items and budget amounts. Data updates are triggered per individual document, and the size of individual documents varies widely, ranging from tens of KB purchase requisitions to hundreds of MB qualification collections. Parsing duration must be adapted to documents of different sizes. Some documents contain non-text content such as supplier qualification scan copies and product parameter images. During parsing, images must be associated with their corresponding business fields to ensure that image-related information can be retrieved during question answering.
How to Configure Settings
The following configuration items are adapted for FastGPT v4.17.0, and can be adjusted for the IT procurement and supplier management scenario.
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Matches the field density of procurement documents, avoiding splitting that damages the integrity of key fields such as purchase items and budget amounts |
chunkSplitter | Split by paragraph + title | Procurement documents mostly have clear purchase requirement and supplier quotation paragraphs. Splitting by title preserves business logic associations |
customPdfParse | Enabled | Most procurement documents are PDF-format contracts and acceptance documents. Enabling custom parsing preserves the original document layout and field structure |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Adapts to the average parsing duration of individual procurement documents, avoiding parsing timeouts for large-volume contracts |
UPLOAD_FILE_MAX_SIZE | 50–100 MB | Matches the common size of qualification documents and quotation attachments included with enterprise procurement orders |
Three Common Misconfigurations
- Phenomenon: After a Word document with images is uploaded to the knowledge base, question answering tests only return plain text fragments and cannot return content related to the images. Cause: The
customPdfParseconfiguration is not enabled, or image links are not associated with their corresponding business fields, and only the plain text content within the document is split. - Phenomenon: After a procurement contract larger than 100 MB is uploaded, the parsing task fails and returns status code 413. Cause: The
UPLOAD_FILE_MAX_SIZEconfiguration value is lower than the actual size of the attachment, exceeding the system's allowed upload limit. - Phenomenon: Search results after chunking include irrelevant non-procurement paragraphs, such as general remarks at the end of the document. Cause: The
chunkSplitterconfiguration does not split by title, and only splits by fixed character length, causing unrelated content across paragraphs to be merged into the same chunk.
How to Verify Proper Configuration
- A standard procurement contract PDF is uploaded, and the parsed chunked content is reviewed to confirm that each chunk contains complete key fields such as purchase items and budget amounts.
- A quotation document with supplier qualification images is uploaded, and the parsing results are checked for association information between image links and their corresponding supplier qualification fields.
- A 50 MB qualification scan collection is uploaded, and the parsing process is confirmed to not trigger timeout errors and complete normally.
chunkSizeis adjusted to a value range that matches the field density of business documents, a small purchase requisition is uploaded, and the chunk length is verified to meet expectations, with no key fields split.
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 2026-10-19.