Build a Multi-Tenant Accelerator Chargeback and Idle Waste Auditor
Correlate allocated GPU tenancy windows with active kernel compute metrics to audit idle capacity costs. Construct transparent chargeback adjustments and team reclaim notices using hosted Claude.
What you are building
The core problem, expected build, and operating context for this challenge.
Correlate allocated GPU tenancy windows with active kernel compute metrics to audit idle capacity costs. Construct transparent chargeback adjustments and team reclaim notices using hosted Claude.
AI Infrastructure Operations Lead
Collect multi-tenant job scheduling logs alongside high-watermark GPU memory and tensor core telemetry.
Isolate reservations where memory was reserved without active compute execution.
Prompt hosted Claude to synthesize an executive chargeback audit explaining preventable idle expenditure.
Deliver an interactive reconciliation report with itemized penalty and credit recommendations.
Executive audit console showing departmental cost attribution, wasted dollar metrics, and appeal workflows.
Generate synthetic JSON event logs mapping pod tenant tags to DCGM compute activity records. Starter examples below are synthetic. They are not downloaded records.
NVIDIA DCGM exporter Extracts GPU tensor core utilization and active memory footprint metrics defined in DCGM exporter.
mlflow, anthropic-claude, pydantic-ai. mlflow tracks multi-run financial experiments, chargeback rules, and cost-recovery metric logs. Hosted Claude API for stakeholder communication and policy audit summaries.
Submit runnable source, pinned dependencies, setup commands, a working interface, example outputs, and a recorded end-to-end demo. Include tests and an action log.
Distinguish between genuine compute idle time and model initialization periods.
Calculate unrecovered financial spend based on per-accelerator hourly internal cost models.
Provide verifiable audit traces mapping job IDs to specific wasted capacity intervals.
```json [ { "input": { "tenant_id": "analytics-org", "billing_rate_per_gpu_hr": 4, "jobs": [ { "job_id": "job-891", "allocated_gpus": 8, "duration_hours": 10, "active_compute_hours": 1.5 } ] }, "expected_output": { "tenant_id": "analytics-org", "total_charged_usd": 320, "idle_waste_usd": 272, "waste_percentage": 85, "audit_ruling": "Job reserved 8 GPUs for 10 hours with 8.5 hours of unutilized memory holding.", "recommended_action": "Apply idle hold surcharge of $272.00 or enforce auto-reclaim after 30 minutes inactivity." } }, { "input": { "tenant_id": "recommendations-team", "billing_rate_per_gpu_hr": 4, "jobs": [ { "job_id": "job-104", "allocated_gpus": 4, "duration_hours": 5, "active_compute_hours": 4.8 } ] }, "expected_output": { "tenant_id": "recommendations-team", "total_charged_usd": 80, "idle_waste_usd": 3.2, "waste_percentage": 4, "audit_ruling": "High efficiency usage observed; exceeds 90% active utilization target.", "recommended_action": "Approve standard chargeback without penalties." } } ] ```
Analytical accounting and billing reconciliation only; produces no real ledger modifications. Use mock external writes. Require human review before any real deployment.
Participation guide
Analytical accounting and billing reconciliation only; produces no real ledger modifications. Use mock external writes and human review.
How work is evaluated
Distinguish between genuine compute idle time and model initialization periods. Calculate unrecovered financial spend based on per-accelerator hourly internal cost models. Provide verifiable audit traces mapping job IDs to specific wasted capacity intervals.
Runnable source, pinned dependencies, setup commands, working interface, example outputs, recorded demo, tests, and action logs.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
Complete workflow
Collect multi-tenant job scheduling logs alongside high-watermark GPU memory and tensor core telemetry. Isolate reservations where memory was reserved without active compute execution. Prompt hosted Claude to synthesize an executive chargeback audit explaining preventable idle expenditure. Deliver an interactive reconciliation report with itemized penalty and credit recommendations.
Missing or incorrect.
Partly complete, with documented gaps.
Complete, with reproducible evidence.
Correctness and evidence
Distinguish between genuine compute idle time and model initialization periods. Calculate unrecovered financial spend based on per-accelerator hourly internal cost models. Provide verifiable audit traces mapping job IDs to specific wasted capacity intervals.
Missing or incorrect.
Partly complete, with documented gaps.
Complete, with reproducible evidence.
Review interface
Executive audit console showing departmental cost attribution, wasted dollar metrics, and appeal workflows.
Missing or incorrect.
Partly complete, with documented gaps.
Complete, with reproducible evidence.
Safe execution
Analytical accounting and billing reconciliation only; produces no real ledger modifications.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Collect multi-tenant job scheduling logs alongside high-watermark GPU memory and tensor core telemetry.
Isolate reservations where memory was reserved without active compute execution.
Prompt hosted Claude to synthesize an executive chargeback audit explaining preventable idle expenditure.
Deliver an interactive reconciliation report with itemized penalty and credit recommendations.
Reference links and supporting material
Generate synthetic JSON event logs mapping pod tenant tags to DCGM compute activity records. Supplied starter examples are synthetic, not source extracts.
Challenge setup
Configuration summary for this challenge.
Challenge input
Input details are not configured.
MLflow
MLOps
Evaluated output
No output checks are configured.
- Collect multi-tenant job scheduling logs alongside high-watermark GPU memory and tensor core telemetry. Iso...
- Distinguish between genuine compute idle time and model initialization periods. Calculate unrecovered finan...
- Executive audit console showing departmental cost attribution, wasted dollar metrics, and appeal workflows.
- 2 public reference cases
- 1 configured tool
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- MLflow · Optional
Evaluation contract
- Complete workflow · Weight 40
- Correctness and evidence · Weight 30
- Review interface · Weight 20
- Safe execution · Weight 10
Reference cases
- 2 public reference cases
Workflow tasks
No workflow tasks are configured.
Recipe freeze marker
No recipe freeze time is recorded.