What the data for this category looks like
This category covers documents including procurement plans, requirement specifications, bidding documents, bid evaluation records, contracts, acceptance receipts, supplier complaint materials, and more. Data sources include enterprise procurement platforms, public resource trading platforms, supplier databases, contract management systems, and audit problem databases. The update frequency is daily.
Documents contain structured fields such as procurement method, budget basis, scoring method, and supplier qualification conditions, alongside unstructured content like review opinions and complaint descriptions. The length of individual documents varies widely, and some bid evaluation records and complaint materials include fragmented segments.
What constraints these characteristics impose on the vector models and indexes workflow
Multi-source data access requires indexes to support multiple data source connections and incremental synchronization, to prevent repeated imports of the same document. The daily update frequency demands a lightweight incremental indexing process, to avoid excessive resource usage from full index reconstruction.
Diverse document types and wide length ranges require chunking strategies tailored to different content lengths, to avoid splitting critical compliance clauses such as procurement methods and scoring methods. Mixed structured fields and unstructured content require indexes to support both field-level matching and semantic matching.
Documents related to audit accountability require high-confidence matching results, so matching thresholds must be strictly controlled to reduce false positive matching risks.
How to configure settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Adapt to the typical content length of most tendering and procurement documents, avoid splitting key compliance clauses |
| Overlap Length | 100–200 characters | Retain overlapping content between adjacent chunks, ensure key compliance information is not cut off during chunking |
| Similarity Threshold | 0.75–0.85 | Calibrated via real-world testing, balance recall rate and accuracy for compliance reviews, reduce irrelevant matching results |
| Single Search Limit | Top 10 entries | Adapt to standard retrieval needs for state-owned enterprise and public sector procurement documents, avoid returning excessive redundant content |
defaultToken | 4096 | Match the context length requirements of most tendering documents, ensure associated documents can be fully utilized during retrieval |
maxToken | 8192 | Support complete indexing and retrieval of long documents, cover extended content such as bid evaluation records and complaint materials |
Three common configuration errors
- Phenomenon: FastGPT returns a rejection when calling a vector model accessed via Ollama, while curl testing works normally. Cause: Correct vector model interface address or authentication parameters are not configured, or interface call frequency limits are not adjusted.
- Phenomenon: Direct management of vector database creation, deletion and other operations via MongoDB is not possible, only the root user has relevant permissions. Cause: The synchronization mapping configuration between FastGPT and the external database is not enabled, or database operation permissions for non-root users are not granted.
- Phenomenon: Duplicate bid evaluation records or supplier information appears after knowledge base index merging, with no deduplication applied. Cause: The index deduplication switch is not enabled, or chunking rules create fragment differences in duplicate content that are too large for the deduplication logic to identify.
How to verify proper configuration
- Import a single complete bidding document, confirm that chunked fragments fully retain key compliance information such as procurement method and budget basis, with no obvious splitting.
- Submit a compliance query for procurement methods, verify that the number of returned matching documents matches the single search limit setting.
- Add a new supplier complaint material via an external MongoDB tool, check if the FastGPT knowledge base synchronizes and updates the new content.
- Import two highly similar bid evaluation records, verify that the merged index automatically filters duplicate entries.
The values provided here 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.