Proactive Sentinel
Primary contributor — Flask backend platform
Multi-tenant SOC platform: event ingestion, MITRE ATT&CK correlation, automated response, and POPIA-aware data handling. Deployed to Railway for evaluation; currently paused.
Architecture
Problem
Financial-services providers need security monitoring that catches multi-stage attacks and business-signal anomalies without a heavyweight enterprise SIEM.
What I built
Events route through a DetectionEngine (Redis dedup + MITRE correlation). After an alert is persisted, an Action Dispatcher fires by severity: critical alerts block the attacker IP in Redis and write an audit entry; lower severities degrade gracefully. Observability is OpenTelemetry plus structured JSON logs.
Engineering decisions
- 01
SHA-256 fingerprint deduplication in Redis suppresses duplicate events at ingest, before they can create alert noise.
- 02
MITRE ATT&CK chain correlation across events — recon followed by lateral movement followed by exfiltration escalates to a critical chain alert, which single-event rules would miss.
- 03
Dispatcher failures are non-fatal: a Redis outage degrades response, never alert persistence.
- 04
POPIA compliance at the data layer: SA and Zimbabwean IDs masked before storage, PII columns encrypted with AES-256-GCM.
Outcome
v1.0.0. Deployed to Railway for evaluation with PostgreSQL 15 and Redis 7; currently paused. 44 passing tests.
