Context
A personal project combining a React data workspace, a TypeScript agent gateway and a Spring Boot data service. It turns natural-language questions into queries that can be confirmed, executed and audited, and connects to a real Oracle 10g database.
Workflow
- Natural language
- QueryPlan
- User confirmation
- Read-only SQL
Implementation
Plan before querying
Parsed natural language into structured QueryPlans and executed queries through a business semantic layer and approved SQL templates. Requested clarification when needed and user confirmation before execution.
Bind confirmation to the actual plan
Used a plan hash to bind user confirmation to the execution plan. Added read-only queries and protection for sensitive fields.
Break down complex questions
Decomposed questions into multiple template queries and calculation steps, consolidated evidence and exposed parsing, execution and analysis progress through SSE.
Keep results traceable
Provided query audit logs and CSV export, and distinguished real data from mock results for review.
Delivery and scope
This personal engineering project connects a real Oracle 10g database to natural-language queries and a frontend analysis workspace. Read-only execution, confirmation and audit logs constrain data access.
Dianhai Wang · Updated