What the data for this category looks like
The internal office data for general contracting of building construction projects comes primarily from project construction drawings, construction logs, material ledgers, labor management documents, internal safety specifications, and OA workflow files. The data update schedule is adjusted based on project progress. In the early project phase, drawings and plans are updated frequently. During the construction phase, progress and material data are updated weekly. Document formats include long-text construction plans spanning dozens of pages, structured bills of quantities, scattered site notices, and labor forms. Fields include project number, building number, construction section, and material model. Units include standard engineering units such as square meters, cubic meters, tons, and man-days.
What Constraints These Characteristics Impose on Knowledge Base Retrieval and Recall
The general contracting of building construction projects has two key characteristics: a large volume of long texts and a high proportion of structured data. This requires knowledge base chunking to balance semantic integrity and retrieval accuracy. Chunks that are too long will introduce redundant context during retrieval. Chunks that are too short will break the logical chain of construction plans. Structured data such as bills of quantities and progress sheets must support field-level retrieval to accurately match information like material models and construction sections. Frequently updated on-site data requires the recall link to support incremental synchronization configuration, preventing retrieval of expired material inventory or progress data. Scenarios with dense professional terms need to use engineering domain word vector models, reducing recall errors for specialized terminology.
Configuration Settings
These configurations apply to FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Match the semantic matching accuracy of engineering terms, avoid recalling irrelevant non-engineering documents |
limit | Top 8–12 entries | Document volume for building construction projects is large per batch, limiting the number of recalled entries prevents context overflow |
searchMode | hybrid | Combine keyword matching and semantic retrieval, cover structured fields and natural language query scenarios |
usingReRank | Enabled | In scenarios with many long texts and professional terms, re-ranking can improve the sorting priority of highly relevant documents |
datasetSearchUsingExtensionQuery | Enabled | For users' vague queries, automatically supplement synonymous terms in the engineering field, improve recall coverage |
collectionIds | Bind specified collections by project | Isolate documents from different projects, avoid interference from cross-project retrieval |
Three Common Misconfigurations
- Symptom: Chinese file names appear garbled when uploading knowledge base files via API. Cause: The correct character encoding format is not specified in the API request header, leading to incorrect parsing of Chinese file names during transmission.
- Symptom: The locally deployed ChatGLM2 model is not available for selection when creating a knowledge base, with no corresponding option in the interface. Cause: The model address and permissions for OneAPI access are not correctly configured in the FastGPT configuration file, so the model fails to register with the platform.
- Symptom: Reference documents in Notion link format cannot be imported. Cause: The adaptation configuration for the corresponding data source is not enabled, so the platform does not recognize the content crawling rules for Notion links.
How to Verify Successful Configuration
- Upload a single engineering long document, run a retrieval test, and check the semantic relevance and field matching of the recall results.
- View the knowledge base synchronization logs to confirm that incremental update tasks run normally according to the preset schedule.
- Check the character encoding configuration of API requests, and verify that Chinese file names display correctly after upload.
- Enter the model management interface, confirm that the locally deployed engineering domain model is correctly loaded and appears in the optional list.
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.