What the data for this category looks like
Meeting minute data sources primarily include meeting recording transcript text, scanned handwritten on-site notes, exported online meeting subtitle files, and digitized versions of paper records compiled by attendees. Updates trigger after each single meeting, with no fixed cycle. One independent document is generated per meeting. Typical document structure includes five fixed modules: meeting topic, attendee list, topic discussion content, resolution items, and follow-up task items. Some documents include attachments such as attendee sign-in sheet screenshots and recording clip links. Fields include attendee name and department, task responsible person and deadline, and unit-based information such as meeting duration and number of attendees.
Constraints imposed on the "document parsing and chunking" link
The fixed module structure of meeting minutes requires that parsing and chunking must not separate topics from their corresponding resolutions and task items. Otherwise, subsequent retrieval cannot associate complete information. Data sources include files in multiple formats. Some files contain embedded images or tables, so the parsing link must support structured extraction and cross-format compatibility. Spoken transcript text has non-standard sentence breaks. Chunking rules must balance coherence and information integrity, and avoid splitting complete discussion paragraphs into overly small units. The length of single meeting minutes varies widely, from hundreds to thousands of characters. Chunking configurations must adapt to documents of different lengths, avoiding overly short chunks that cause information fragmentation, or overly long chunks that cause context redundancy.
How to set the configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Meeting minutes contain coherent topic discussions and resolution items. This length preserves complete information for a single topic and avoids separating key associated content |
chunkSplitter | Split by paragraph + title | Meeting minutes have clear topic and module titles. Splitting by paragraph combined with titles preserves module integrity |
customPdfParse | Enable structured parsing | Some meeting minutes are submitted in PDF format, with embedded tables or images. Structured parsing preserves table fields and image-related text |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Single meeting minutes may include multiple attachments. This upper limit covers most enterprise-level meeting total file sizes |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Large meeting minutes contain long text segments and multiple attachments. This duration prevents task failure due to parsing timeout |
chunkSettingMode | Manual configuration | The module structure of meeting minutes varies widely. Manual configuration allows flexible adjustment of chunking rules to adapt to different minute formats |
Three Common Mistakes
- When uploading a meeting minute file with embedded images, the parsed result does not include text descriptions or annotations corresponding to the images. The symptom is that the parsed text has no content related to images, and the file parsing status shows success but lacks information. The cause is that the image content extraction configuration in structured parsing is not enabled, or the dependency modules for image parsing are not loaded.
- After importing a meeting minute in Excel format, the chunking result loses the row and column association information of the table. The symptom is that table content is split into scattered single-line text, and structured data such as original attendees and task items cannot be restored. The cause is that no dedicated chunking rule is configured for table content, and only the default character-by-character splitting logic is used.
- A parsing timeout error occurs when parsing a minute file generated from a long meeting. The symptom is that the system returns a timeout error message, and the parsing task is forcibly terminated. The cause is that the configured value of
PARSE_FILE_TIMEOUT_SECONDSis lower than the actual required parsing duration, and cannot cover the parsing process of large files.
How to Confirm the Configuration Is Correct
- Upload a standard-format meeting minute test file, and check whether the parsed chunking result preserves the integrity of core modules such as topics and resolutions.
- Check the file size and timeout records in the parsing log, and confirm that the configured values of
UPLOAD_FILE_MAX_SIZEandPARSE_FILE_TIMEOUT_SECONDSmatch the actual situation of the test file. - For meeting minute files containing tables or embedded images, verify that the parsed result preserves table fields and text annotations related to images.
- Adjust the configured value of
chunkSize, compare the chunking results under different values, and confirm that the chunking length adapts to the retrieval requirements of the current business.
The values provided here are common starting points, and should be evaluated 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.