PeopleOS
A local-first People Intelligence system that separates verified workforce calculations from optional AI interpretation.
- Python
- People Analytics
- Local-first
Under the hood
Technical architecture
Runs locally with a Next.js frontend and a Python API bound to loopback by default. Source availability does not imply a hosted demo or a production certification.
Workspace
Next.js frontend
Supports import review, questions, cohort exploration and progressive disclosure of evidence.
Application boundary
Python API · FastAPI
Validates dataset meaning and requests, enforces role boundaries when exposed remotely, and governs analysis and model lifecycle actions.
Evidence
Deterministic analytical engines
Computes workforce metrics and governed cohort results with denominators, exclusions, provenance and minimum group sizes.
Optional interpretation
Local Ollama
Organizes verified results into explanations. Citations, scope and policy are checked server-side; the model does not own the calculations.
Failure and fallback path
When local AI is unavailable, deterministic evidence remains available. Missing or ambiguous evidence keeps the relevant capability unavailable instead of fabricating a result.
One request, end to end
- Import a test dataset, review field mappings and confirm meaning.
- Run governed analytics and produce typed evidence with provenance.
- Inspect that evidence directly or request an optional local AI explanation.
Why it is built this way
- Local-first processing limits unnecessary workforce-data transfers.
- Privacy thresholds constrain aggregate drill-down.
- Predictive activation has a separate governed lifecycle; asking a question does not activate a model.