Customer feedback ingestion pipeline
Store.AI / IRA
Weekly-refreshed ingestion of 8.3M+ customer feedback records from three sources into DuckDB, explorable from region down to a single outlet.
(Python · Gemini · DuckDB · Streamlit · pandas · SQL)
# Problem
Customer feedback for a large restaurant chain lived in three separate sources: post-order surveys, delivery ratings, and public reviews. Each had its own rules for which responses count, so the same question gave different answers depending on where you looked.
# Approach
Built one ingestion pipeline that pulls all three sources, tags free-text feedback with an LLM, and lands everything in embedded DuckDB while keeping each source’s counting rules intact. Data loads weekly because the volume keeps growing. A Streamlit dashboard on top drills from region to city to a single outlet.
# Result
One place to read 8.3M+ feedback records (7.8M + 382K + 142K across the three sources), refreshed every week, from national view down to one outlet.