The JD states "Cloud-based data platforms (AWS today, GCP future)" — GCP fluency signals you are hired for where Carrier is going, not just where it is. TEKsystems lists Google Cloud as a named strategic partner.
Google Cloud Platform (GCP) is Google's public cloud. For AI/ML workloads, the central platform is Vertex AI — Google's fully managed ML platform that unifies the entire lifecycle: data prep, training, evaluation, deployment, monitoring, and GenAI. Think of it as GCP's equivalent to AWS SageMaker, but with Gemini (Google's foundation model family) natively integrated.
The key insight for Carrier: The architectural patterns are cloud-agnostic. Every AWS service has a direct GCP counterpart. The migration from AWS to GCP is a service substitution, not an architectural redesign. The data pipeline, feature store, model registry, and monitoring patterns John knows from AWS translate directly — different service names, same architecture.
John's GCP credential: Google Cloud Fundamentals certification. Entry-level but it signals intentional engagement with the platform — not just AWS-only thinking.
| GCP Service | What it is | Carrier use case |
|---|---|---|
| Vertex AI | Unified ML platform: training, evaluation, deployment, monitoring, GenAI — single UI and API surface | Future replacement for SageMaker across Carrier's AI/ML fleet |
| Vertex AI Pipelines | Kubeflow-based ML pipeline orchestration — DAG of data prep → train → eval → deploy steps | HVAC predictive maintenance retraining pipeline |
| Vertex AI Feature Store | Centralized feature storage — online (low-latency serving) and offline (batch training) | Shared HVAC sensor features across models and BUs |
| Vertex AI Model Monitoring | Automated drift detection for skew and drift in production model inputs/outputs | Alert when HVAC equipment fleet generation shifts sensor distributions |
| Vertex AI Agent Builder | Managed platform for building RAG apps and agentic AI — grounding, retrieval, tool use. Equivalent to Bedrock Agents + Knowledge Bases. | Future "Tell Me More" GenAI agent migration from Bedrock to GCP |
| Gemini | Google's foundation model family (Ultra, Pro, Flash, Nano). Available via Vertex AI — equivalent to Claude on Bedrock. | Foundation model for Carrier's GenAI applications on GCP |
| BigQuery | Serverless data warehouse with lakehouse capabilities — petabyte scale, built-in ML (BigQuery ML), native vector search | Future destination for Carrier's HVAC sensor data lake and analytics |
| BigQuery ML | Run ML models directly in BigQuery SQL — train, evaluate, predict without moving data out of the warehouse | Rapid prototyping of fault prediction models on historical HVAC data |
| Pub/Sub | Managed real-time messaging — equivalent to Kinesis. Sub-second latency for IoT event streams. | Real-time HVAC sensor ingest pipeline on GCP |
| Dataflow | Fully managed stream and batch data processing (Apache Beam). Equivalent to Glue for ETL at scale. | ETL from raw IoT sensor data to BigQuery features |
IoT ingest: Edge devices → Pub/Sub → Cloud Storage (raw) → Dataflow ETL → BigQuery (features) → Vertex AI Feature Store
Traditional ML: Feature Store → Vertex AI Training Job → Model Registry → Vertex AI Endpoints → Model Monitoring → alerts
GenAI: Technical docs → Cloud Storage → Vertex AI Search (vector index) → Agent Builder → Abound API → technician chat
Governance: Vertex AI Explainability + Model Monitoring (drift) + Cloud Audit Logs + IAM + VPC Service Controls
RAG index rebuild risk: Bedrock Knowledge Bases use S3 and OpenSearch under the hood. Vertex AI Agent Builder uses Cloud Storage and Vertex AI Search. The RAG index rebuild is the highest-risk step in the migration — embedding model parity and retrieval quality need evaluation gates before cutover.
"I'd build the evaluation harness first — BERTScore against ground-truth Q&A pairs on the existing Bedrock Knowledge Base, then run the same eval against the rebuilt Vertex AI Search index. Only promote when retrieval quality is within tolerance."
| MLOps Capability | GCP Native | John's Equivalent |
|---|---|---|
| Pipeline orchestration | Vertex AI Pipelines (Kubeflow) | SageMaker Pipelines (Amgen), DeepEval CI/CD (NewsRx) |
| Model registry & versioning | Vertex AI Model Registry | MLflow at NewsRx |
| Feature management | Vertex AI Feature Store | SageMaker Feature Store (Amgen) |
| Production tracing | Cloud Logging, Cloud Trace | Langfuse self-hosted at NewsRx |
| Drift detection | Vertex AI Model Monitoring | HHEM + BERTScore trending; SageMaker Model Monitor |
| GenAI platform | Agent Builder + Gemini | vLLM + Qwen3 (self-hosted); Bedrock (managed equivalent) |
| Data warehouse | BigQuery | Redshift (J&J MedTech), cross-client Snowflake |
| Streaming ETL | Pub/Sub + Dataflow | Kinesis + Glue (AWS); GeoSpatial Metrics UAV telemetry |
| Evaluation gates | Vertex AI Model Evaluation | DeepEval + HHEM + BERTScore CI/CD gates (NewsRx) |
| Canary / A-B deployment | Vertex AI Endpoints (traffic split) | Staged facility-by-facility rollout (Invistics) |
TEKsystems' email footer lists Google Cloud as a named strategic partner alongside AWS, Microsoft, Red Hat, Tableau, and Snowflake. This is not a speculative future — TEKsystems is actively positioning Carrier's GCP migration as part of this engagement. The person hired for this role will be expected to architect the transition.