OpenAI Agents SDK Hardware Serving Profiler & AWS Neuron Router
Build a hardware-aware serving performance profiler using OpenAI Agents SDK to analyze token generation latency and power consumption across AWS Neuron and CUDA runtimes. Implement automated request routing logic that optimizes throughput per watt, achieving a 25% cost reduction under SLA latency constraints.
What you are building
The core problem, expected build, and operating context for this challenge.
Utilize OpenAI Agents SDK to profile and route inference requests dynamically between AWS Neuron (Trainium/Inferentia) and NVIDIA CUDA accelerators.
How work is evaluated
Evaluates cost savings and SLA adherence for hardware request routing.
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.
sla_compliance
Ensures achieved P99 latency remains below SLA limit.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
cost_savings_pct
Percentage savings compared to 100% CUDA deployment. • target: 25 • range: 0-60
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Build an OpenAI Agents SDK autonomous routing agent for hardware runtime selection
Profile AWS Neuron (NeuronCore-v2) vs NVIDIA CUDA (H100) performance profiles across batch sizes
Implement dynamic SLA-aware request dispatching based on real-time queue latency and power efficiency
Achieve 25%+ total inference cloud cost reduction while meeting 99th percentile latency SLAs
Reference links and supporting material
Telemetry dataset covering 10,000 inference requests executed on AWS inf2/trn1 instances versus H100 GPU nodes with recorded latencies and pricing.
Challenge setup
Evaluates cost savings and SLA adherence for hardware request routing.
Challenge input
JSON request batch with prompt token counts, output target lengths, and SLA latency bounds.
OpenAI Agents SDK
Official framework for constructing dynamic task routing agents.
Evaluated output
JSON routing breakdown (Neuron % vs CUDA %), average P99 latency, and overall cost savings %.
- Ensures achieved P99 latency remains below SLA limit.
- Percentage savings compared to 100% CUDA deployment. • target: 25 • range: 0-60
- Benchmark: LLM Serving Cost-Efficiency Benchmark
- Cost Savings Pct target: 25
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
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
- OpenAI · Required
- Anthropic · Optional
- Azure OpenAI Service · Optional
Evaluation contract
- sla_compliance · Weight 1
- cost_savings_pct · Weight 1
Metrics
- cost_savings_pct · Target 25
Benchmarks
- LLM Serving Cost-Efficiency Benchmark
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON request batch with prompt token counts, output target lengths, and SLA latency bounds.
Output: JSON routing breakdown (Neuron % vs CUDA %), average P99 latency, and overall cost savings %.
Execution harness: Python
Evaluation tools
- OpenAI Agents SDK · Agent Framework
Recipe freeze marker
No recipe freeze time is recorded.