Build a GPU Reservation and Team Cost Reconciliation Engine
Cross-reference multi-tenant GPU reservation allocations against DCGM active telemetry to pinpoint unused reserved capacity. Formulate team chargeback adjustments and idle reclamation actions for infrastructure finance teams.
What you are building
Cross-reference multi-tenant GPU reservation allocations against DCGM active telemetry to pinpoint unused reserved capacity. Formulate team chargeback adjustments and idle reclamation actions for infrastructure finance teams.
Infrastructure Financial Operations Lead
Ingest scheduled GPU instance reservations and hourly team lease cost agreements.
Parse normalized DCGM GPU utilization telemetry records to compute actual hardware occupancy.
Calculate unallocated idle waste and formulate reconciled departmental cost attribution line items.
Invoke hosted Claude via Anthropic Python SDK to synthesize a finance reconciliation memo.
FastAPI service with a Streamlit cost reconciliation audit table and markdown report export.
Generate 20 normalized synthetic hourly GPU reservation records and matched DCGM average utilization export logs. Starter examples below are synthetic. They are not downloaded records.
NVIDIA DCGM exporter Extract DCGM exporter metric concepts like GPU utilization and active tenant profiling from the DCGM exporter overview.
anthropic, pydantic, fastapi. anthropic-claude structures departmental variance explanations into audit-ready reconciliation memos for infrastructure leadership. Hosted Claude API accessed via Anthropic Python SDK with structured JSON parsing capabilities.
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.
Correctly identifies reserved GPU hours where DCGM compute utilization fell below 5 percent.
Applies chargeback recalculations strictly within agreed hourly rate parameters.
Exports structured JSON reconciliation entries accompanied by a Claude-drafted audit memo.
```json [ { "input": { "team_id": "search-ranking", "reserved_gpus": 8, "reserved_hours": 24, "hourly_rate": 3.5, "avg_dcgm_utilization_pct": 2.1 }, "expected_output": { "allocated_cost": 672, "effective_used_cost": 14.11, "idle_waste_reclaimable": 657.89, "status": "underutilized_reclaim_recommended", "action_code": "DOWNSIZE_RESERVATION" } }, { "input": { "team_id": "core-nlp", "reserved_gpus": 4, "reserved_hours": 12, "hourly_rate": 3.5, "avg_dcgm_utilization_pct": 0 }, "expected_output": { "allocated_cost": 168, "effective_used_cost": 0, "idle_waste_reclaimable": 168, "status": "dormant_lease", "action_code": "CANCEL_RESERVATION" } } ] ```
Simulates financial cost chargebacks and resource adjustments without modifying actual cloud billing accounts or instance states. Use mock external writes. Require human review before any real deployment.
Rules
Simulates financial cost chargebacks and resource adjustments without modifying actual cloud billing accounts or instance states. Use mock external writes and human review.
How work is evaluated
Correctly identifies reserved GPU hours where DCGM compute utilization fell below 5 percent. Applies chargeback recalculations strictly within agreed hourly rate parameters. Exports structured JSON reconciliation entries accompanied by a Claude-drafted audit memo.
Runnable source, pinned dependencies, setup commands, working interface, example outputs, recorded demo, tests, and action logs.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
Complete workflow
Ingest scheduled GPU instance reservations and hourly team lease cost agreements. Parse normalized DCGM GPU utilization telemetry records to compute actual hardware occupancy. Calculate unallocated idle waste and formulate reconciled departmental cost attribution line items. Invoke hosted Claude via Anthropic Python SDK to synthesize a finance reconciliation memo.
Missing or incorrect.
Partly complete, with documented gaps.
Complete, with reproducible evidence.
Correctness and evidence
Correctly identifies reserved GPU hours where DCGM compute utilization fell below 5 percent. Applies chargeback recalculations strictly within agreed hourly rate parameters. Exports structured JSON reconciliation entries accompanied by a Claude-drafted audit memo.
Missing or incorrect.
Partly complete, with documented gaps.
Complete, with reproducible evidence.
Review interface
FastAPI service with a Streamlit cost reconciliation audit table and markdown report export.
Missing or incorrect.
Partly complete, with documented gaps.
Complete, with reproducible evidence.
Safe execution
Simulates financial cost chargebacks and resource adjustments without modifying actual cloud billing accounts or instance states.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Ingest scheduled GPU instance reservations and hourly team lease cost agreements.
Parse normalized DCGM GPU utilization telemetry records to compute actual hardware occupancy.
Calculate unallocated idle waste and formulate reconciled departmental cost attribution line items.
Invoke hosted Claude via Anthropic Python SDK to synthesize a finance reconciliation memo.
Resources
Generate 20 normalized synthetic hourly GPU reservation records and matched DCGM average utilization export logs. Supplied starter examples are synthetic, not source extracts.
Challenge setup
Challenge input
Input details are not configured.
Anthropic Claude
LLM for code generation & more
Evaluated output
No output checks are configured.
- Ingest scheduled GPU instance reservations and hourly team lease cost agreements. Parse normalized DCGM GPU...
- Correctly identifies reserved GPU hours where DCGM compute utilization fell below 5 percent. Applies charge...
- FastAPI service with a Streamlit cost reconciliation audit table and markdown report export.
- 2 public reference cases
- 1 configured tool
Shows the first task, up to three tools, and up to three evidence items.
View technical recipe
Configured tools
- Anthropic Claude · 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.