Sandbox & AI Toolkits
Sealed experimentation environments in Google Cloud Malaysia: evaluate, compare and retrain without data leaving the platform boundary.
Model test-bench
Evaluate any catalogue model against a governed dataset or your own CSV, inside a sealed sandbox.
Side-by-side comparison
Run one input through two models and compare outputs, confidence, and latency in parallel.
No-code retrain
Pick a registered model, point it at fresh lakehouse data and launch a governed Vertex AI retraining run.
Notebook environment
Vertex AI Workbench notebooks with governed BigQuery datasets and the Vertex AI SDK ready to use.
Compute environments
| Environment | Specification | Status | Indicative chargeback |
|---|---|---|---|
| CPU · Standard | n2-standard-8 · 8 vCPU · 32 GB | Available | Included in BU budget |
| GPU · Shared L4 | g2-standard-8 · 1× L4 · 24 GB | Available | MYR 6.80 / hour |
| GPU · A100 | a2-highgpu-1g · 1× A100 40 GB | 2 in queue | MYR 21 / hour |
| Training cluster | a2-highgpu-8g · 8× A100 40 GB | By reservation | MYR 165 / hour |
All sandboxes are network-isolated inside VPC Service Controls: no internet egress, governed datasets read-only, outputs scanned before export. Compute is charged back to your business unit's cost centre in MYR. Sessions auto-terminate after 8 idle hours.