What the data for this category looks like
The core users of this category are project managers, implementation managers, and PMOs. The documents processed daily include four types: project plans, milestone lists, weekly reports, and risk issue lists. Data sources cover project contracts, work breakdown structures (WBS), client schedule arrangements, and content exported from project management systems. The update frequency is once per week or per milestone node. Document structures vary significantly, ranging from pure text descriptions to structured documents with embedded tables, progress nodes, and embedded images. Core fields include task name, responsible person, start and end time, risk level, and progress percentage.
What constraints these characteristics impose on the "document parsing and chunking" link
Document sources include contracts, WBS, client schedules, and project management system content. Formats include mixed structures of plain text, tables, and embedded images. Parsing logic must adapt to multiple data source types and formats. Documents contain bound fields such as tasks, responsible persons, and deadlines. When chunking, the association between fields and their corresponding content must be retained to avoid damaging business logic after splitting. The update frequency is once per week or per milestone. A single document may contain multiple pages. The parsing process must have reasonable timeout thresholds and batch processing capabilities. Chunking configurations must flexibly adapt to project documents of different structures.
How to set the configurations
The following configurations are formulated based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Documents in this category mostly contain short paragraph task descriptions and long text progress descriptions. This range can retain complete associated information for a single task or single risk item |
chunkSplitter | Split by title + paragraph | Documents in this category have hierarchical structures such as project plan titles and task node titles. Splitting by title avoids damaging business logic associations |
customPdfParse | Enabled | Most documents in this category are exported PDF formats with embedded tables and images. Enabling custom parsing retains table structures and image association information |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | A single project document may contain multiple pages of WBS or weekly report content. This duration covers the complete parsing process |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Bulk imported project document packages may reach a relatively large volume. This upper limit covers conventional bulk import needs |
chunkSettingMode | Manual configuration | The structure of documents in this category varies significantly. Manual configuration flexibly adapts to different types of project documents |
Three common mistakes
- Phenomenon: Uploading a Yuque public share link prompts parsing failure. Cause: The link was not confirmed to be fully publicly accessible. FastGPT only supports parsing public web data sources that meet standard formats. Unauthorized permission restrictions block parsing requests.
- Phenomenon: Embedded images are not extracted from the parsed PDF document. Cause: The
customPdfParseconfiguration was not enabled. The default parsing logic cannot capture PDF embedded images and convert them into recognizable content, only extracting the plain text part. - Phenomenon: The association information between tasks and responsible persons is lost after chunking. Cause: The
chunkSplitterconfiguration was set to forced splitting by fixed character length, without anchoring title or paragraph boundaries. This splits the core fields of a single task into different content chunks.
How to confirm the configuration is correct
- Upload a project weekly report PDF that contains embedded tables and images. Check whether the parsing result retains the table structure and annotation information of embedded images.
- View the running logs of the parsing task. Confirm that the parsing time does not exceed the configured value of
PARSE_FILE_TIMEOUT_SECONDS, and there are no timeout-related error messages. - Randomly extract content fragments after chunking. Verify that a single chunk contains complete associated fields such as task name, responsible person, and deadline.
- Upload a Yuque public share link of a project document. Confirm that parsing is successful and the content is complete, with no permission-related error prompts.
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.