Three production incidents diagnosed from first principles:

 81% error rate under load  →  uvicorn --workers is anti-pattern under K8s
                                (shared CPU budget = thrashing, not parallelism)
                                Fixed: asyncio + ThreadPoolExecutor, GIL analysis
                                Result: 81% errors → 0%, 2000m CPU → 1000m

 SHAP returning all zeros   →  TreeExplainer incompatible with StackingClassifier
                                Fixed: KernelExplainer in original feature space
                                Evaluated 4 alternatives before deciding

 HPA never scales down      →  Memory-based HPA + fixed ML footprint
                                = mathematically impossible to scale down
                                Fixed: CPU-only HPA, 3→1 pods in 8 minutes

I spent 14 years running business operations — teams, vendors, budgets and customer-facing processes — before moving into ML engineering. That background is why my portfolio looks the way it does: a model is useful only when it can be deployed, monitored, explained, improved, and operated with cost discipline.

I build working ML systems, document trade-offs, measure failures, and turn those lessons into reusable patterns. Seeking my first formal ML/MLOps role; experienced in ownership, pressure and making systems easier for the next person to operate.

Target roles: ML Engineer · MLOps / Production ML · Platform Engineer (ML infra) · ML Platform / Data Engineering with ML workflows — evaluated on evidence, not tenure.

Is a reusable foundation for teams that want safer defaults when moving machine learning services toward production.

I created it after building my portfolio and seeing the same failure patterns repeat: blocked APIs, fragile deployments, missing monitoring, unclear model promotion rules, weak secrets handling, and documentation that does not match how the system actually behaves.

The template packages those lessons into a reusable starting point:

# Zero to working fraud detection service in one command
git clone https://github.com/DuqueOM/ml-service-template.git
cd ml-service-template && make bootstrap

→ Template repo  |  QUICK_START.md  |  43 ADRs

The template's philosophy generalized to local, multi-tier LLM agents — not a fork, a reusable platform (core/ + thin usecases/ domains) that proves the governance model travels beyond tabular ML serving.

→ agent-local repo  |  Full write-up

Shows three end-to-end ML projects built beyond notebooks: model training, APIs, containers, deployment artifacts, monitoring, tests, and documented engineering decisions.

One example from the portfolio: a load test exposed an 81% error rate in an ML API. I traced the issue to a serving pattern that created CPU contention under Kubernetes, then redesigned the inference path with asynchronous execution and a thread pool. The result dropped the error rate to 0% and reduced CPU needs.

That story matters because it shows the habit I want to bring to an ML/MLOps role: measure the problem, understand the cause, fix it, and document the lesson.

📐 18 ADRs →  |  📋 Engineering Highlights →  |  📺 3min Demo →

Python scikit-learn XGBoost LightGBM PySpark FastAPI Docker Kubernetes Terraform GitHub Actions MLflow DVC Prometheus Grafana SHAP AWS GCP

TripleTen Data Science
14 years operations -> MLOps & Production ML

I use AI-assisted coding tools — and I engineer that workflow instead of hiding it. My production template encodes the governance layer: behavior protocols (AUTO/CONSULT/STOP), path-scoped rules, an append-only audit trail and eval gates that keep AI-generated changes reviewable and bounded. Architecture, trade-off analysis, incident diagnosis, and final technical decisions are my responsibility; the tooling accelerates the rest.

I consider this a core engineering skill for 2026, not a disclaimer.