This document is written based on the configuration logic of FastGPT v4.17.0.
What the Data for This Category Looks Like
Data sources include engineering orders, sand, gravel, cement admixture inspection records, moisture content test reports, mixing plant control system data, and laboratory test records. Daily generated documents cover mix proportion notices, opening identification forms, production and transport orders, raw material inspection reports, test block compressive strength records, and more. Data is updated per vehicle production record, per batch raw material inspection, and daily laboratory ledger updates. Document fields include raw material batch number, moisture content value, batching ratio parameters, pouring location, strength grade, and other items. Corresponding units include kilograms, cubic meters, megapascals, and others.
How These Characteristics Impact Knowledge Base Retrieval and Recall
The high-frequency updated production and inspection data for this category requires the knowledge base retrieval link to support incremental synchronization and expired data filtering. This prevents the use of expired raw material mix ratios or transport records. Document fields include strongly business-related items such as raw material batches, moisture content, and strength grades. Retrieval must use precise field matching to ensure recalled results align with the engineering conditions of the current bidding project. Single production records have strong timeliness, so the recall range must be limited to valid data from the past 6 months. This avoids mix ratio parameters becoming invalid due to raw material batch changes. In bidding scenarios, historical quotations with the same strength grade and same pouring location must be associated. The retrieval logic must add business tag filtering to narrow the scope of invalid recalls.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirements for core business parameters of ready-mixed concrete, avoiding recall of irrelevant historical records |
limit | Top 8 entries | Balances retrieval efficiency and recall coverage, avoiding excessive redundant data from occupying the context window |
usingReRank | Enabled | Corrects mix ratio parameter misalignment that may occur with pure semantic retrieval, and strengthens business relevance ranking |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands business query terms related to bidding, covering associated dimensions such as mix ratios, transport, and strength |
collectionIds | Bind collections of this enterprise’s raw material inspection, production records, and historical bidding documents | Limits retrieval scope to this enterprise’s valid business data, avoiding interference from cross-entity data |
Three Common Misconfigurations
- Phenomenon: Semantic retrieval scores are mostly above 0.9, and recalled results include a large number of irrelevant non-concrete business documents. Cause: No reasonable
similaritythreshold is set, or no exclusive business collection is bound, leading to an overly large retrieval scope. - Phenomenon: After enabling
usingReRank, only 1 result is finally returned for quotation generation. Cause: The result interception rule after re-ranking is incorrectly configured, truncating all re-ranked results to a single entry, resulting in loss of multi-dimensional mix ratio and quotation reference data. - Phenomenon: Single retrieval time exceeds 15 seconds, and the page loads without response. Cause: No retrieval collection scope is limited, or no reasonable
limitparameter is set, causing the system to scan excessive expired production and inspection data.
How to Confirm Proper Configuration
- A test query containing a specific strength grade (such as C30) is uploaded, and the retrieved results are checked to confirm they only include this enterprise’s concrete business documents.
- After enabling
usingReRank, the retrieval result list is viewed to confirm the number of returned results is greater than 1, and that the sorting conforms to business relevance. - A batch of test Excel raw material inspection data is imported, and all fields are confirmed to be correctly identified and indexed.
- A single retrieval request is simulated, and the time consumption is checked to confirm it is within a reasonable range, with no timeout occurring.
The values provided here are common starting points, and should be evaluated against the user’s own samples.