Training & Retraining

The accuracy of the DBN Analytics POC API depends entirely on the quality of its training data stored in ChromaDB. This section explains everything you need to know about the RAG training lifecycle — from initial setup to ongoing refinement.

• Concepts

  Understand how RAG training works and why it improves SQL accuracy.

• Initial Training

  Bootstrap the vector store from your database schema and first SQL examples.

• Retraining

  Add, inspect, and remove training data using the training API endpoints.

• Best Practices

  Guidelines for writing high-quality training data that maximises accuracy.

---

Training API Endpoints at a Glance

Method	Endpoint	Description
GET	/api/v1/training/data	List all training records
POST	/api/v1/training/ddl	Add a DDL (table schema)
POST	/api/v1/training/sql	Add a question–SQL pair
POST	/api/v1/training/documentation	Add a business documentation string
POST	/api/v1/training/from-database	Auto-train from the live database schema
DELETE	/api/v1/training/{id}	Remove a training record by ID