Tools I’ve worked with

DATA ENGINEERING

Pipelines and data quality

Python and SQL sit at the center. I use DuckDB, Parquet and PyArrow for local analytical processing, and PostgreSQL and dbt for relational modeling and tested transformations.

• Python
• SQL
• DuckDB
• Parquet / PyArrow
• PostgreSQL
• dbt

• Validation and deduplication across historical event partitions.
• Restartable monthly Gold builds and bounded aggregation.
• Project work with Redpanda and PySpark Structured Streaming.

MACHINE LEARNING & EVALUATION

Model training and evaluation

I use scikit-learn and PyTorch to build models and MLflow to track experiments. I check chronological splits, compare against a baseline, and report what changed on the holdout.

• scikit-learn
• PyTorch
• MLflow
• pandas / NumPy
• pytest
• FastAPI

• Chronological splits and train-only preprocessing.
• Baseline comparisons, fixed seeds and a separate final holdout.
• Ranking metrics, forecast errors, model limitations and API serving.

DELIVERY & OPERATIONS LABS

Deployment and recovery

My platform experience comes from independent labs. Terraform provisions GCP/GKE; GitHub Actions builds and promotes; Argo CD reconciles the workload.

• Docker
• GCP / GKE
• Terraform
• GitHub Actions
• Helm / Argo CD
• Prometheus / Grafana

• Keyless OIDC / Workload Identity Federation and exact image digests.
• API health and metrics, alert routing, and recovery exercises.
• Linux and Bash throughout the lab workflow.

The verified GKE scope is the API health, metrics and delivery path. The full data and ML stack was not established as running in GKE.