SignalHR
SignalHR is an HR agent on Snowflake + Claude.
Core trick
Scores team burnout risk from pager, after-hours and time-off data under Snowflake access policies, with a second model verifying each number.
Description
SignalHR is an HR agent on Snowflake + Claude. It spots teams at risk of burnout and the promises we haven’t kept, approves time off without breaking on-call, and prepares calibration. Snowflake policies govern access, a second Claude verifies every number, and humans approve every action.
README excerpt
# Ingestion · SignalHR on Snowflake **Find which teams are about to burn out or break, and what we promised them, before anyone resigns.** The signals that predict regretted attrition are spread across systems: pager logs, after-hours work, time not taken, pulse surveys, reviews and goals. The promises managers make in 1:1s and retros sit in notes nobody tracks. **Demo video (1 min):** [docs/people-guard-demo.mp4](docs/people-guard-demo.mp4) (recorded under the working title "People Guard") **SignalHR** is an agent for HR Business Partners and people managers, connected to **Snowflake** and powered by **Claude**. It covers three cases: - **Teams at risk:** which teams are at risk of losing people, and what did we promise them? - **Time off:** which pending requests can we approve without breaking on-call coverage, and which comp days do we owe? - **Performance reviews:** where should the next calibration look closer: rating mix, self vs manager gaps, completion? > **Revenue version: on hold.** The repo also contains a Revenue version (renewals at risk and AI token consumption for a VP of Customer Success). Its code, data and Snowflake objects are kept, but the app offers only the HR version by default. Set `ENABLED_EDITIONS=hr,revenue` to bring it back. The sections below still describe both. For every case the agent does the same five things: 1. **Find:** queries Snowflake under the user's own role (semantic views, health views, raw tables). 2. **Explain:** combines t