Side Project2024

NIQQueryAgent

Ask questions about syndicated data in plain English

PythonOpenAINIQ/CircanaSQL

Overview

Natural language interface to query syndicated data (NIQ/Circana). Ask questions in plain English and get data-backed answers. Eliminates weeks of analyst time for routine data pulls.

Key Metrics

Natural Language
Interface
NIQ/Circana
Data
Weeks → Minutes
Time Saved

Case Study

The Problem

Querying syndicated data (NIQ/Circana) requires specialist knowledge of complex data schemas and hours of analyst time for routine pulls. Brand managers wanting to answer simple questions like "how is our share trending?" had to wait days for analyst support.

The Approach

1

Built a natural language interface layer that translates plain English business questions into structured NIQ/Circana queries

2

Engineered prompts to handle ambiguous business questions — mapping intent like "Southeast performance" to the correct geographic hierarchy

3

Added a data validation layer that catches misinterpretations before surfacing results, so outputs can be trusted

4

Eliminated any need to know the data schema or query syntax — users just ask in the language of their business

The Impact

Reduced routine data pulls from weeks of analyst time to minutes of self-service querying

Unlocked self-service analytics for brand managers who previously had no path to syndicated data without analyst support

Demonstrated the repeatable pattern for natural language querying of any structured, schema-heavy data source

Lessons Learned

"The hardest engineering challenge isn't NL parsing — it's reliably mapping ambiguous business intent to precise data schema concepts"