I'm a software engineer and cybersecurity researcher based in Nairobi, Kenya. My work sits at the intersection of backend engineering, ML infrastructure, and applied security — I build systems that matter at scale, from distributed infrastructure to models in production.

I care about the operational reality of software: what happens after deployment, how systems behave under failure, and how to make complex distributed infrastructure understandable to the people who run it.

Production-grade ML monitoring infrastructure. Not a notebook. Not a demo.

ML models silently degrade when real-world data distributions shift away from training data. A model can have 99.9% uptime and zero HTTP errors while producing systematically wrong predictions for real users. I built a system that catches this automatically.

A fully containerised, self-healing ML monitoring system that detects distribution shift in live inference traffic within minutes and responds autonomously — no human intervention required.

Statistical detection layer — three complementary methods, each catching what the others miss:

Infrastructure highlights:

• Kafka-backed inference interceptor adds < 2ms latency to every prediction request
• Faust-Streaming processor computes drift scores on tumbling windows (1,000 records or 60 seconds)
• Redis-backed distributed circuit breaker — state shared across replicas, never in-process memory
• Alert deduplication via atomic SET NX EX — prevents alert storms across multiple alerter replicas
• Airflow DAG fetches fresh data, retrains, validates F1 against production model, and promotes via MLflow registry — fully automated
• Auto baseline registration: new model versions self-register reference distributions from live traffic
• FastAPI dashboard with JWT auth, WebSocket live updates, asyncio-safe broadcast with heartbeat/ping
• 13-panel Grafana observability suite with Prometheus scraping all services
• Kubernetes-ready: HPA, NetworkPolicy, PodDisruptionBudget, zero-downtime rolling deployments

Security: OWASP headers · JWT + bcrypt (factor 12) · SlowAPI + Redis rate limiting · Parameterised SQL · SASL/SCRAM-512 Kafka · TLS PostgreSQL · Non-root containers · RBAC · Full audit logging · OpenTelemetry tracing.

Stack: Python 3.11 FastAPI Kafka Faust-Streaming PostgreSQL 16 Redis 7 Airflow MLflow Prometheus Grafana Docker Kubernetes NumPy SciPy OpenTelemetry

• Deepening expertise in distributed systems and production ML infrastructure
• Learning Rust for systems-level programming
• Researching applied cryptography and network security
• Building in public and representing the African engineering community globally