apps

NSSA Multi-Agent Report Generator

AI BI platform for NSSA that turns natural language into governed SQL, reusable templates, and interactive reports with admin review built in.

Repository

PythonSQLLLMsFastAPIPostgreSQLStreamlit

Deep dive

Case Study

Problem

NSSA teams needed self-serve insights from the Employers database without SQL skills or manual report backlogs. Ad hoc questions were slow, hard to reuse, and risky to run unchecked.

Approach

Built a multi-agent NL-to-SQL platform with a template-first strategy and human-in-the-loop approval: - Orchestrator routes intent (report / analysis / generic) - Vector search reuses approved SQL templates before generating new queries - RAG (Ollama + schema context) drafts SQL when no template exists - Admins review, edit, and approve queries before production use - Analysis agent adds insights, charts, and trend detection on top of results

Results

Stakeholders query employer data in plain language and get tables, exports, and visualizations in one flow. Template reuse cuts repeat LLM calls; admin review and SQL validation build trust. Streaming progress, chat history, and feedback loops improve matching over time.

Learnings

Production-ready NL-to-SQL needs governance, not just generation — template libraries, admin approval, and layered validation matter as much as the model. Intent routing and follow-up handling (e.g. “show me a graph”) keep UX fast and costs down.