Compensation analytics dashboard

// the problem

Compensation questions ('average total comp for female physicians in Texas vs. male physicians') are easy to ask in plain English and hard to answer without a real analytics layer underneath.

// the approach

• –Built 6 dashboard views (overview, pay by role, geography, pay equity, FT/PT parity, merit planning) over ~1,200 synthetic employee records.
• –Chat interface where Claude translates a plain-English question into a strict query spec, validated against a schema whitelist.
• –The query itself runs as an exact pandas computation — Claude never does the math, only the translation and the write-up of results already computed.
• –Deliberately used a seeded, deterministic data generator instead of prompting Claude for synthetic records, and a grouped bar chart instead of a choropleth map, to keep the demo fast and dependency-light.

// the outcome

6 working dashboard views plus a chat interface that answers ambiguous natural-language questions with numbers that are always exactly right, because Claude never touches the arithmetic.