People analytics · Interactive org chart · LLM assistant
BetaOrg Health Explorer
An open-data rebuild of an organizational analytics product.
Loading the company…
About this tool
A working model of the kind of tool I build for HR and business leaders: span of control, management layers, co-location, attrition, and office attendance for a realistic synthetic company. Drill through the organization as a sunburst, treemap or org chart, and ask an assistant questions about it.
The Org Health Tool I launched at Uber runs on confidential employee data, so I can't show it. This project recreates the approach on Kestrel, a fictional 5,000-person company with three years of history. A layoff, a return-to-office mandate and a delayering are written into the data, and the detection models find each one without being told where to look.
How it works
- 01
Generate a realistic company
A Python simulation builds reporting lines, levels and locations, then runs 36 months of hires, exits, promotions and office attendance, with known events planted in the history.
- 02
Measure org health
One bottom-up pass per month rolls every metric up to every manager, so any leader's organization has a full history: span of control, layers from the CEO, manager co-location, voluntary and involuntary attrition, attendance and tenure.
- 03
Detect what changed
Change point detection finds lasting shifts, a rolling-median filter finds one-month anomalies, and each finding lists the sub-organizations that moved most.
- 04
Explore the structure
A sunburst, a treemap and an org chart, all drillable. Group by department or location, size by headcount, and shade by span of control or average tenure.
- 05
Ask the assistant
Claude answers plain-language questions by calling a fixed set of metric functions. It never sees raw records, and the server refuses any request outside the organization being viewed.
Next project
Inflection