Problem
What made this worth building.
When anyone has a question about data, they need either a dashboard (if it exists) or an analyst. Both are constrained. Dashboards are static and sometimes poorly built. Analysts are scarce. I wanted to remove that dependency for routine business questions. Most questions center around "what happened?" — movements in KPIs and metrics. The second layer is "why did it happen?" Seeing how much bandwidth goes into answering these two questions repeatedly, I decided to build something that bypasses dashboards and SQL knowledge entirely.
Value
What the app gives back.
Data Whisperer is built on KPI contracts and semantic layers. These define what metrics mean, how they're calculated, where the data comes from. The LLM never guesses — it knows what "revenue" means because it's defined somewhere and is instructed to strictly refer to the definition and a specific table. The first question is always operational: what happened? How did the KPIs move? What changed versus yesterday, versus last week, versus last year? But that's just the start. The second question is the real one: why did it change? Contribution analysis answers that — it breaks down the change across grouped attributes. Which customer segment moved the needle, which product line, which geography. That narrows the investigation. That forms the next set of questions.
