NTelligence

Team Name

NTelligence

Team Members

Project Description

NTelligence is a conversational analytics tool for NT Government. Staff ask plain-English questions (e.g., “Top 5 departments by average rating in 2023”) and NTelligence returns a safe, explainable answer with evidence: the query plan, parameterised SQL, parameters, rows, and a grounded summary.
It reduces spreadsheet/SQL chaos, speeds up decisions in regional and central teams, and bakes in privacy controls (allow-listed tables/columns, date windows, aggregation/masking for sensitive attributes).

How it works (one-liner):
UI → FastAPI (server.py) → Agent 1 Scope Guard → Agent 2 Planner (QueryPlan JSON) → Agent 3 SQL (parameterised) ↔ KB (tables + curated views) → Agent 4 Verify → answer + evidence.

Built-in demo prompts:

“Top 5 departments by avg rating in 2023”

“Staff allocation next week — Darwin & Palmerston”

“Scaffold queries” (shows Plan → SQL → Rows → Summary)

Data Story

Data story (Markdown supported)

Problem
Across agencies, routine questions require ad-hoc SQL and scattered spreadsheets. That’s slow, inconsistent, and risky around sensitive data—especially for small/regional teams.

Data approach

Demo KB: synthetic employee, action, perf tables + curated views joinempperf, joinempaction.

Execution: Agent 3 compiles parameterised SQL and runs only on allowed tables/views.

Protection: Scope Guard enforces allow-lists, clamps date windows, and requires aggregation/masking for sensitive fields where policy demands.

Verification: Agent 4 grounds summaries only in returned rows and surfaces simple trust signals (e.g., coverage, freshness).

Why this helps the NT

Speed & fairness: regional offices get the same analytical power as central teams—less waiting, more consistent answers.

Evidence-based allocation: staffing and programs can be directed using fresh, comparable metrics.

Transparency: every answer includes the plan, SQL, params, and rows—easy to check or reproduce.

What we built (hack deliverables)

Backend (FastAPI/uvicorn, Pydantic v2, pydantic-ai, SQLAlchemy, SQLite demo; Postgres-ready)

System architecture diagram + sequence diagram

Suggested-prompts UI flow (shows Scaffold queries: Plan → SQL → Rows → Summary)

README + seed DB for instant run

Limitations & next steps

Add charting agent (auto-visualise rows), caching/cost controls, and connectors for common NT data stores & the NT Open Data Portal.

Expand policies per-agency; incorporate more detailed freshness/quality checks.