Amin Al-Ait
Cloud Data Engineer with 4+ years of hands-on AWS experience, most of it on serverless data pipelines for automotive test data: proprietary formats like ATFx, ATFx-NVH, MF4 and Modal alongside standard tabular data. Contributed to all seven of the platform's ETL Terraform modules.
Helped build the data platform for an automotive enterprise, cutting analysis costs by over 50% and time-to-result by over 75%. Applies a full Testing Pyramid and privacy-aware GenAI tooling to every engagement. Also builds QGI, a personal applied-ML platform on AWS that ranks countries by their resemblance to the historical trajectories behind past crises.
- Contributed to all seven serverless ETL pipelines (AWS Lambda, Python, Terraform) of an automotive data platform, covering formats ATFx, ATFx-NVH, MF4 and Modal. Cut analysis cost >50%, time-to-result >75%.
- Ran the benchmarks the architecture was chosen on: a 175-run InfluxDB vs Athena comparison (Athena ingestion 4 to 6 times faster) and a 20-variant Parquet encoding study on latency and cost.
- Owned the ETL testing strategy across the full pyramid: pytest with mocking and coverage gates, custom stub gateways, black-box application tests, and golden-file end-to-end runs locally and against live AWS.
- Managed infrastructure via Terraform across AWS Lambda and ECS, and migrated the production Grafana estate across demo, QA and prod over InfluxDB Flux, Athena SQL and Prometheus.
- Led a proof of concept digitising semi-structured test reports with AWS Textract, replacing its unreliable query mode with full form extraction, a custom Python parser and human-in-the-loop review.
- Authored the internal DiTiCS Python library (MF4 processing, typed, test-covered behind pre-commit hooks) and the team's dependency-vetting checklist and Docker base-image vetting process. Worked with LLM coding tools daily on a privacy-compliant model.
- Built and owned the complete serverless backend for a mobile running coaching application on AWS Lambda, API Gateway and Aurora.
- Implemented GPS data analysis to assess athlete pace and geolocation performance.
Live, solo-built serverless AWS platform across 7 repositories. Reads a country’s multi-decade indicator trajectory against the historical record and surfaces the past cases it most resembles. 250 indicators from 8 providers, 10,459 verified events (1903 to 2026), 71 categories, 214 countries.
Per-event-type model fleet. 42 independently trained binary classifiers, each with its own greedy-selected indicator set, lookback window searched across nine horizons, and architecture choice.
Validation and integrity engineering. Leave-one-country-out cross-validation, bias-corrected bootstrap intervals, time-matched negative sampling, and pre-registered gates that catch its own false positives.
Platform. Multi-account AWS org with Terraform state in S3, GitHub Actions OIDC and no long-lived credentials; Step Functions over AWS Batch on Graviton Spot; a 14.3-billion-row Parquet lake on S3, Glue and Athena; a Next.js dashboard served from baked static JSON.
Presented at the 8th PPE Conference, University of Witten/Herdecke.