What the data for this category looks like
Core data for survey and design comes from full-cycle project survey reports, official design drawing instructions, industry standard atlases, project ledgers, and change records. Data update frequency varies by project phase. Update frequency is higher during the scheme demonstration phase, and relatively stable during the construction drawing phase. Latest version content is synchronized when industry standards are updated. Document structures include long text paragraphs (single reports can span dozens of pages), structured metadata such as project number, survey point coordinates, standard document numbers, and metadata associated with drawings. Fields use engineering units including length, area, load, and others.
What constraints these characteristics impose on the knowledge base retrieval and recall link
The high proportion of long text requires retrieval to retain contextual connections between chunks, to avoid breaking the complete semantics of professional terms. Structured metadata requires retrieval to support precise filtering by fields such as project number and standard document number, to reduce interference from cross-project documents. The irregular update rhythm requires the knowledge base to support incremental updates, ensuring that the latest content after project changes can be quickly retrieved. Metadata associated with drawings requires recall results to synchronously link to corresponding drawing instructions, improving information completeness.
How to set up configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.85 | Survey and design documents contain many professional terms. A threshold that is too low introduces irrelevant standard content, while a threshold that is too high misses relevant design descriptions from the same specialty. Calibrated based on actual testing. |
limit | Top 8–12 entries | Single projects have large document volumes. Too many results exceed the context window, while too few miss critical design parameters. |
searchMode | vector + keyword | Balances semantic matching for professional terms and exact matching for standard numbers and project numbers, adapting to the mixed retrieval needs of survey and design documents. |
usingReRank | Enabled | Semantic drift often occurs after long text is chunked. Reranking corrects the initial recall order and improves the ranking of relevant results. |
datasetSearchUsingExtensionQuery | Enabled | Survey and design queries often carry implicit needs. For example, asking about foundation treatment automatically links to geotechnical survey standards. Extended queries cover more relevant documents. |
collectionIds | Configured by project group | Prevents interference from cross-project documents, ensuring retrieval results only come from the currently assigned survey and design projects. |
Three common mistakes
- An API call passing knowledge base-related parameters returns a parameter error and empty fields. Cause: The
collectionIdsparameter is not configured correctly, or the global variable is not bound to the knowledge base collection ID for the corresponding project. - When uploading a single large PDF survey report, progress reaches 90% and then prompts "offset out of range". Cause: Large file chunked upload logic is not adapted, or the chunk size exceeds the single-block processing limit allowed by the system.
- Retrieval results recall survey standards from unrelated projects, and the number of results does not match expectations. Cause:
collectionIdsis not configured by project group, or thesimilaritythreshold is set too low, introducing irrelevant cross-project documents.
How to confirm the configuration is correct
- Call the API with the
collectionIdsparameter, then check if returned results only include survey and design documents for the current project. - Upload a single PDF survey report larger than 100 MB, and confirm there are no errors after upload progress completes.
- Enter a professional query such as "foundation bearing capacity design parameters", then check if recall results include the corresponding standards and survey data within the project.
- After enabling
usingReRank, compare the initial recall order with the reranked order, and confirm relevant documents are ranked higher.
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.