Muhammad Husnain
I design and operate ingestion, orchestration, and warehousing systems for marketing analytics, geospatial ML pipelines, and SaaS data products. 6+ years across five companies. 200+ production pipelines shipped. Comfortable owning the layer from API ingestion through dbt models to Terraform-managed infrastructure.
I work at the boundary between data engineering and infrastructure — the part where pipelines need to be reliable, cheap to run, and observable by the people downstream from them.
At Making Science I led the build-out of a multi-channel marketing platform pulling from 15+ ad APIs into per-client BigQuery projects. The architecture was hub-and-spoke: a shared ingestion layer feeding isolated customer warehouses, orchestrated by Airflow on Cloud Composer, transformed through layered dbt models. The interesting problems were rarely about ingestion itself — they were about cost, tenancy, and standing up new customers without engineering toil. The Terraform module library I built brought new-pipeline deployment from an hour to roughly fifteen minutes.
Before that I worked on geospatial ML pipelines at LiveEO (satellite imagery, Anyscale, spot EC2), and AWS warehousing at Ryan-Miranda. My early years were spent building a low-code data platform on Flask, Nifi, and a stack of custom connectors — which is where I learned that the boring parts of data work (auth, retries, schema drift, idempotency) are usually the ones that matter most.
note Currently studying for the GCP Associate Data Practitioner cert and open to senior data engineering roles — remote, or relocation to the right team.
Hub-and-spoke BigQuery architecture ingesting 15+ ad APIs into per-client projects. Cloud Functions and Pub/Sub for serverless ingestion, Cloud Composer for orchestration, dbt for the transformation layer. Serves 400+ clients with cadences down to 15 minutes.
Company-wide modules for 8 GCP services (Pub/Sub, Cloud Functions, Cloud Run, Cloud Scheduler, BigQuery, IAM, Secret Manager, GCS). Standardized naming, IAM, and observability. Cut new-pipeline deployment from ~1 hour to 10–15 minutes.
Migrated 20+ customers from Dataproc clusters to Cloud Functions, removing always-on cluster cost and improving cold-start economics. Per-customer infra cost dropped ~58% (~$12 → ~$5 per day).
Distributed compute on Anyscale + EC2 spot/on-demand for vegetation, infrastructure, and deforestation models. 20–100+ GB per run. Spot-heavy scheduling cut monthly compute ~80%.
GA4 + Postgres + MySQL ingestion orchestrated by MWAA into Redshift, with Terraform-managed dev and prod environments. Replaced manual spreadsheet reporting across 3 client environments.
Flask backend, drag-and-drop ML integrations, and an Apache Nifi ETL layer that let 20+ client companies compose pipelines without code.
Best way to reach me is email. I read every message and reply within a day or two.