Data EngProof of Work

Multicloud Portability — One Contract, Three Clouds

The same openFDA fact runs on GCP BigQuery, Microsoft Fabric, and AWS serverless — and the counts, business metrics, and schema reconcile exactly. Moving the platform can't change the truth.

View Code

🟰 300

Records match all 3 clouds

🚨 68.33%

Serious rate — identical

🔗 fingerprint

Schema reconciles

💸 <$0.0001

AWS cost (serverless)

Multicloud Portability — One Contract, Three Clouds — demo

Tech Stack

GCP BigQueryMicrosoft FabricOneLakeAWS S3AWS DynamoDBTerraformPythonDelta LakePyArrow

Before / After

Before

Three Clouds, Three Truths

Vendor lock-in risk

GCP result

↓

≠ Fabric result

↓

≠ AWS result

↓

Nobody reconciles

Each cloud computes its own numbers, and a silent drift between them is nobody's job to catch.

Logic is copied per cloud, so a fix on one is a bug on the others.

"Portable" is a slide, not a proven property.

Migrating means hoping the KPIs survive the move.

After

One Contract, Three Native Implementations

Moving can't change the truth

Shared openFDA contract

↓

BigQuery · Fabric · S3+DynamoDB

↓

Each cloud-native

↓

Reconciled exactly

✓

The Same Numbers, Everywhererecord_count, serious_count, serious_rate (68.33%), distinct_drugs, and total_reactions (759) reconcile EXACTLY across GCP, Fabric, and AWS — each computed independently on its own cloud's copy.

✓

Schema Is Enforced, Not AssertedEach cloud's ACTUAL schema is fingerprinted from real data and validated against one shared versioned contract — not a constant copied three times.

✓

AWS Serverless, Not A Server FarmopenFDA lands in S3 (typed Parquet) and serves from DynamoDB (on-demand) — two serverless services, no RDS/OpenSearch/always-on.

✓

Cheap And ReversibleThe whole AWS slice costs under $0.0001 with a documented teardown (--teardown / terraform destroy) and a 30-day S3 expiry.