Problem
Operational teams need to understand whether data is complete, timely, and trustworthy. When health signals are scattered or manual, issues are discovered late and investigation starts without shared context.
Projects · Selected systems
What was built, why it mattered, and what changed.
These are deliberately high-level accounts. They share the problem, my way of thinking, and the lessons without exposing proprietary code, internal tooling, client information, or non-public architecture.
Business observability
DATA QUALITY / OPERATIONS / MONITORING
Approach
Designed monitoring and dashboard workflows around the questions operators actually need answered: what changed, where the gap is, and what requires attention. The emphasis was visibility rather than another opaque layer.
Outcome
Clearer operational visibility and a more consistent basis for identifying data-quality and completeness issues. Specific internal metrics and implementation details remain private.
Lesson
Observability is most useful when it reflects business meaning, not only technical health. A green process can still produce an outcome people cannot trust.
AI productionization
TIME SERIES / HUMAN REVIEW / RELIABILITY
Problem
An AI-assisted prototype can be convincing in isolation while remaining difficult to trust in a repeatable operational workflow. Time-series work adds sensitivity to context, drift, and exceptional conditions.
Approach
Focused on the workflow around the model: structured inputs, review points, traceable outputs, and human judgment where uncertainty matters. Productionization was treated as a systems problem rather than a model-selection exercise.
Outcome
A practical direction for turning AI-assisted analysis into a usable, reviewable process, with guardrails that keep the human operator informed rather than bypassed.
Lesson
The difficult part of applied AI is often not generating an answer. It is creating the evidence, interfaces, and operating model that let someone decide whether to act on it.
Personal homelab
RASPBERRY PI / DOCKER / NETWORKING
Problem
I wanted a secure, flexible environment for learning infrastructure, networking, automation, and self-hosting without abstracting away how the pieces connect.
Approach
Designed and maintain a Raspberry Pi-based environment using Docker, Cloudflare Tunnel, Tailscale, Grafana, Uptime Kuma, Homebridge, AdGuard Home, and small self-hosted utilities.
Outcome
A reliable personal platform that supports experimentation and daily use: private remote access, service monitoring, DNS-level filtering, home automation, and room to test new tools safely.
Lesson
Operating systems yourself teaches lessons tutorials cannot. Failure, recovery, maintenance, and trade-offs become real when the system supports your own routines.