From Chatbot to Detective: Introducing Recursive Insight Generation
MAJOR UPDATE: THE COGNITIVE ENGINE
Lumina is no longer just a tool you talk to. It now talks to you. We've implemented Lumina Radar (Background Exploration) and the Detective Engine (Recursive Reasoning) to automate the discovery of root causes.
In the evolution of the Dialogue Intelligence Framework™ (DIF), we have focused heavily on the Interpretation and Explanation layers, helping the AI understand your questions and answer them clearly.
But there has always been a gap in modern analytics: The questions you don't know to ask.
A dashboard is passive; it waits for you to look at it. A chatbot is reactive; it waits for you to prompt it. Today, we are moving to the Exploration Layer (X-Layer). We are making Lumina proactive.
1. Lumina Radar: The Silent Scout
Imagine hiring a brilliant analyst who, the moment they get a dataset, immediately runs 50 statistical tests before you even walk in the room. That is Lumina Radar.
When you upload data, Radar runs a background sweep using our local DuckDB engine. It isn't just looking for "High Sales"; it is looking for the Physics of your business:
Offset Analysis
Radar detects Simpson's Paradox. For example, if your Global Revenue is flat, Radar will find that "Region A" is crashing while "Region B" is booming, canceling each other out. A standard dashboard would just show "Flat Growth." Radar flags the hidden crisis.
Volatility & Shifts
It calculates rolling standard deviations to find "Step Changes." It knows the difference between a seasonal dip and a fundamental baseline shift that requires attention.
2. The Detective: Recursive Reasoning
Finding an anomaly is easy. Explaining why it happened is hard. This is where most AI tools fail. They hallucinate a reason.
We solved this with the Recursive Detective. This feature resides in the Insight Layer (I-Layer) of DIF. Instead of guessing, the AI enters a loop:
- Signal Capture: Radar finds two facts: "Revenue is down" and "Wait Time is up."
- Hypothesis Formulation: The Detective asks itself: "Is high wait time causing the revenue drop?"
- Proof Generation (The Magic): The AI writes a specific SQL query to test this correlation mathematically.
- Synthesis: If the data proves the link, it generates a "Compound Insight" for the user.
Scenario 1: The Coffee Shop Crisis
We've released a new demo dataset to showcase this: "Radar Test: Coffee Shop Crisis."
In this scenario, a coffee chain looks healthy on the surface. Total revenue is stable. However, Radar automatically detects that the "Downtown" location has crashed by 40%.
Simultaneously, the Detective notices a 400% spike in "Avg_Wait_Time" at that same location. It connects the dots and presents you with a single notification:
"The Revenue Drop in Downtown is directly correlated (0.92) with a spike in Wait Times starting 30 days ago."
Scenario 2: The Retail Stock Out
The same dataset also contains a hidden retail nightmare at the "North Outlet" location.
Revenue hits zero. Waste Cost hits zero. A naive dashboard might show this as "100% Efficiency" (Zero Waste!).
Radar's Shift Detection engine notices the sudden baseline drop in both metrics. The Detective then scans for leading indicators and finds that Inventory_Percent hit 0% exactly 1 day before the revenue crash.
Instead of celebrating efficiency, it alerts you: "Revenue collapse caused by total stock out."
Why Local-First Makes This Possible
Running these kinds of recursive, multi-pass SQL queries on a cloud server would be prohibitively expensive and slow. Because Lumina runs Local-First via WebAssembly, we can fire off hundreds of SQL queries in the background on your device without costing a dime or risking data privacy.
The future of analytics isn't a better dashboard. It's an analyst that never sleeps.
Run Your First Radar Scan
Upload your data and let Lumina's always-on detection engines find the anomalies your dashboards are missing.