All projects

Project breakdown · Early Development

Sentinel

An early-access preview (0.5) for mixed on-premises data centers: operators register equipment by site, room, and rack, connect read-only host or targeted Redfish collectors, and follow telemetry history, threshold alerts, and coverage gaps in one workspace. Security investigations and workstation tools remain available separately.

PythonFastAPIReactTypeScript
Sentinel data-center dashboard preview with sample data

Why I built it

The product started as a single-PC monitor; the work outgrew that shape, so the primary surface became a mixed on-premises data-center control plane.

What works today

What exists today: site, room, and rack inventory, atomic telemetry batches, confirmed threshold alerts with acknowledgement, coverage-gap visibility, a read-only host collector, targeted Redfish snapshots, and a Windows 0.5 early-access preview. SNMPv3, hypervisor and BMS connectors, maintenance workflows, and notification delivery remain planned.

What I wanted to learn

  • Normalize facility telemetry behind one authenticated API.
  • Keep collection read-only and explicit — no discovery, no device actions.
  • Expose coverage gaps and uncertainty instead of certifying health from missing data.

How it works

  1. Register inventory explicitly by site, room, and rack with expected intervals and thresholds.
  2. Accept atomic telemetry batches from authenticated service collectors only.
  3. Confirm threshold transitions only on repeated readings; missing or stale data never clears an alert.

Technical structure

A FastAPI control plane owns org-scoped inventory, telemetry ingestion, alert state, and audit transitions behind role-based auth: admin registration, service-key reporting, dashboard reads. Read-only collectors run inside the site network — a host collector for CPU, RAM, and system-volume usage and a targeted Redfish collector for system, chassis, and sensor snapshots over verified TLS — and push normalized readings to the API. A React and TypeScript dashboard renders inventory, history, alerts, and coverage gaps. The Windows console still runs its owned API on loopback for local preview.

Testing

  • API tests cover tenant boundaries, atomic batches, duplicate and out-of-order handling, and nonfinite-input rejection.
  • Collector tests cover TLS boundaries and identical-observation retries.
  • CI runs API, web, windows-monitor, and semgrep checks on every change.

Deployment

The Windows 0.5 early-access preview is distributed from the public landing page. Remote collectors need a separately deployed reachable HTTPS control plane, and polling physical equipment runs inside the site, not in a serverless request.

Technical decisions

  • Keep collectors read-only: no subnet discovery, no followed links, no device actions.
  • Treat unknown, stale, and failed readings as not healthy — never as proof of health.
  • Keep the preview free with no checkout or paid-plan promise while the product is validated with operator feedback.

Challenges

Combine inconsistent facility telemetry without hiding different meanings, update schedules, or failure modes — and without presenting the preview as a complete DCIM or production monitoring platform.

What I learned

This project is teaching me how to scope a product honestly: what is implemented, what is planned, and what the preview explicitly does not claim — and to keep collection read-only when the tool touches other people's equipment.

What I'd improve next

Validate the preview with operator feedback, then expand connector coverage before making any production claims.