Skip to content
Felipe Sousa

Home Network Observability System

Sentinel: a home-network monitoring and alerting system with device discovery, DNS visibility, category intelligence, risk scoring, and an operator dashboard. Being refactored into its own repository.

ObservabilityBackendDevOps/Infra

Problem

Typical home networks provide poor visibility into device behavior, DNS activity, and operational events, making monitoring and control fragmented and shallow.

Solution

An always-on network sensor feeds device discovery and DNS query events into a shared platform API, which normalizes, categorizes, and scores that data for a dedicated operator dashboard with alerting, thresholds, and reporting.

Architecture highlights

  • Always-on sensor: ARP-based discovery on Linux, ICMP/ARP-cache discovery on Windows, DNS visibility via polling the query log of a local DNS resolver
  • Platform API ingesting devices and DNS events behind a sensor-authenticated boundary, backed by Postgres/Prisma
  • Seven scheduled workers: rollups, offline detection, archival, risk review, threshold alerts, daily email digests, and a health check on the system itself
  • Next.js dashboard with device risk scoring - always shown with the breakdown of the factors that produced it, never an opaque number - category breakdowns, alerts, custom thresholds, CSV export, and on-demand PDF reports
  • Self-health monitoring: the system detects and alerts on its own failure modes (DNS resolver down, sensor stalled, host address changed), checked from the worker rather than the sensor, precisely because the sensor is one of the things being watched
  • Two-factor access even on a surface that is not reachable from the internet - network position is not an authorization model

Challenges

  • Cross-platform device discovery with no single reliable OS API - ARP-based on Linux, ICMP/ARP-cache on Windows
  • Validating the system against a live, real home network instead of synthetic data
  • A router restart broke DNS for an entire household for days, and diagnosing it was a long manual investigation - the system could describe every device on the network and had nothing to say about itself

Responsibilities

  • Define modular architecture for monitoring and control flows
  • Implement DNS and network visibility foundations end to end, from sensor to dashboard
  • Model database entities for devices, sessions, DNS activity, categories, and alerts
  • Ship the operator dashboard, alerting, and reporting capabilities

Outcomes

  • Created strong applied proof beyond ecosystem-only architecture work
  • Demonstrated systems thinking across network, backend, data, and frontend layers
  • Was validated end to end against a live home network, with real device discovery and real DNS traffic, while it ran as a platform module

Learnings

  • Validating against real network and DNS traffic surfaced edge cases synthetic data would not have shown
  • A monitoring system that cannot see its own outage reports "all quiet" and "I am dead" identically. Closing that gap needed no new infrastructure: it reused the alert feed that already existed, with three new device-less alert types
  • Proof of life has to be a direct signal, not an inference: deriving "the sensor is alive" from recent DNS events confuses a stalled sensor with a sleeping household, and turns every night into a false positive

Stack

TypeScriptNext.jsReactNode.jsPostgreSQLPrisma