What the data for this category looks like
Survey and design work consulting data primarily originates from full-project-cycle survey reports, design construction drawing explanations, national and industry standard drawings, project change notices, and similar materials. Data updates follow a per-project incremental model, with uploads or modifications made as needed during project progression. Vector index synchronization typically completes within 10–30 minutes after a single modification. Document structures include long-form technical descriptions (typically 500–5000 characters), structured fields such as project ID, survey point coordinates, design parameters, and standard specification numbers. Common engineering units such as meters, megapascals, and square meters are used. Some documents include OCR text extracted from scanned drawings.
What constraints these characteristics impose on the knowledge base retrieval and recall link
The multi-type features and update patterns of survey and design data create multiple constraints for the retrieval and recall process. Long-form technical descriptions must retain professional parameter integrity, and excessive chunk splitting must be avoided to prevent professional parameters from being split apart. Structured fields must support exact matching to avoid confusing coordinates or bearing capacity parameters across different projects. The per-project incremental update model requires precise retrieval range limiting by project to prevent cross-project data interference. Both short standard specification texts and long technical texts require balanced recall, to ensure coverage of both general standard clauses and project-specific technical details.
How to configure the settings
This content adapts to the knowledge base configuration logic for FastGPT v4.17.0. The specific configuration items and recommendations are as follows:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Survey and design documents contain many professional terms. A threshold that is too low will introduce irrelevant standard clauses, while a threshold that is too high will miss relevant technical description content |
limit | Top 10–15 results | The number of knowledge base documents per project is moderate. Too many will exceed the context window, while too few will miss key chunked technical content |
searchMode | Semantic + exact hybrid mode | Both semantic matching results for professional terms and exact matching content for structured fields must be recalled |
usingReRank | Enabled | Survey and design documents have many similar technical parameter descriptions, and reranking can improve the ranking priority of highly relevant content |
datasetSearchUsingExtensionQuery | Enabled | Expanding the wording of professional terms can match more accurate standard clauses and project technical descriptions |
collectionIds | Specified as an array format using project IDs | Survey and design work is managed by project classification, and limiting the recall range can avoid confusion from cross-project data |
Three common configuration mistakes
- Issue: The knowledge base retrieval interface call returns a parameter format error, with the error prompt "invalid collection id". Cause: The
collectionIdsparameter is not passed in array format, and a single string-type knowledge base ID is passed directly. - Issue: Retrieval results do not include the core content of the target PDF document, and the number of results is far lower than expected. Cause:
datasetSearchUsingExtensionQueryis not enabled, andsearchModeis only set to exact mode, which cannot match the long-text chunked content in the PDF document. - Issue: The exported knowledge base file does not include recently modified documents. Cause: The export operation runs before waiting 10–30 minutes for vector index updates to complete, and modified documents have not yet completed knowledge base ingestion.
How to confirm the configuration is correct
- Input a test question containing professional terms, such as "the design value of vertical compressive bearing capacity of a project's pile foundation", and check whether recall results include the corresponding project's survey report content.
- Review the ranking of retrieval results, confirm that highly relevant content appears in the top 3 entries, to verify that the
usingReRankconfiguration is active. - Pass the specified
collectionIdsarray parameter, and check whether recall results only include knowledge base documents for the corresponding project, with no cross-project data included. - Test passing a variable-form
collectionIdsparameter, confirm that no errors occur during parameter parsing during the conversation process, to verify correct variable adaptation.
The values provided here are common starting points, and should be evaluated against the relevant 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.