Jetson Edge Intrusion Detection

Technical Brief

Current CSV-driven analytics, planned Jetson flow ingestion, stable event contracts, and defensive boundaries.

Open dashboard | README | Architecture | Thor runbook

Technical Brief

Jetson Edge Intrusion Detection is built around one architecture rule: adapters may change, but the analytics pipeline should not. The current implementation validates the lookback, forecasting, alert, and reporting layers with fixed CSV telemetry as a deterministic fixture.

Current vs Planned Pipeline

LayerCurrent working systemPlanned Jetson ingestion upgrade
Input sourceFixed CSV fixtureJetson-generated flow CSV
Capture modeDeterministic replaySPAN, TAP, or local interface capture
Packet stageNot required for current evidenceRotating PCAP files
Flow extractionCSV columns normalized into TelemetryEventZeek conn.log, Suricata eve.json, CICFlow-style records
Analytics pathLookback analytics, forecasting, alerts, reportsSame existing analytics path
Dashboard impactImplementedNo detector/dashboard rewrite intended
Thor benchmarkMeasured run committedMeasured inference run on jetsonthor (tegra264); CPUExecutionProvider

Architecture Principle

Adapters may change. The analytics pipeline should not.

The planned upgrade adds a Jetson packet/flow ingestion stage before the existing CSV contract. New sources should normalize into the same event/schema boundary so the detector, lookback, forecasting, alerting, and dashboard layers remain stable.

Adapter Status

AdapterStatusPurpose
CsvTrafficSourceImplemented / current fixtureReads fixed CSV telemetry and emits normalized events.
ZeekConnLogSourcePlannedNormalize Zeek conn.log records into the event contract.
SuricataEveJsonSourcePlannedNormalize Suricata eve.json flow and alert records.
CicFlowCsvSourcePlannedNormalize CICFlow-style records.
PcapFlowSource / PcapCaptureStagePlannedCapture or replay packets, rotate PCAP files, and feed defensive flow extraction.

Boundary