Build an Accelerator Procurement Capacity and Cost Simulation Engine — implementation

## Build
Model multi-accelerator cloud procurement strategies against projected generative AI inference demands. Evaluate cost tradeoffs between spot, on-demand, and reserved capacity under preemption risk.

## User
Cloud FinOps Architect

## Workflow
1. Ingest projected hourly token request demand and model parameter specifications.
2. Simulate provisioning across AWS Neuron instances and NVIDIA GPU cloud offerings.
3. Prompt hosted Claude to evaluate risk-adjusted cost matrices and spot preemption exposure.
4. Generate a procurement portfolio report with cost savings and availability guarantees.

## Interface
Scenario comparison dashboard displaying cost curves, preemption vulnerability graphs, and allocation tables.

## Data
Construct synthetic hourly demand profiles and pricing tables for accelerated EC2 instances (Inf2, Trn1, and GPU equivalents).
Starter examples below are synthetic. They are not downloaded records.

## Source
[AWS Neuron documentation](https://raw.githubusercontent.com/aws-neuron/aws-neuron-sdk/master/README.md)
Utilizes AWS Neuron accelerated EC2 instance family specifications (Inf1, Inf2, Trn1, Trn2) for architectural cost modeling.

## Tools
pydantic-ai, anthropic-claude, mlflow.
Enforces strict financial schema validation and constraint checking on simulated procurement allocations.
Hosted Claude API for narrative tradeoff analysis between reservation commitments and spot preemption costs.

## Deliverables
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.

## Acceptance
- Simulates multi-tier procurement models satisfying peak hourly token demand.
- Quantifies potential cost savings and preemption interruption probabilities.
- Outputs structured JSON procurement allocations conforming to schema guidelines.

## Starter Examples
```json
[
  {
    "input": {
      "monthly_demand_tokens_billion": 50,
      "target_model": "Llama-70B",
      "latency_sla_ms": 100,
      "budget_limit_usd": 25000
    },
    "expected_output": {
      "recommended_fleet": [
        {
          "instance_type": "trn1.32xlarge",
          "pricing_model": "1yr_reserved",
          "count": 4,
          "monthly_cost_usd": 14200
        },
        {
          "instance_type": "inf2.48xlarge",
          "pricing_model": "spot_flexible",
          "count": 2,
          "monthly_cost_usd": 4800
        }
      ],
      "total_monthly_usd": 19000,
      "projected_sla_conformance_pct": 99.4,
      "risk_commentary": "Reserved Trn1 provides steady baseline; spot Inf2 absorbs morning bursts"
    }
  },
  {
    "input": {
      "monthly_demand_tokens_billion": 1,
      "target_model": "Qwen-7B",
      "latency_sla_ms": 500,
      "budget_limit_usd": 2000
    },
    "expected_output": {
      "recommended_fleet": [
        {
          "instance_type": "inf2.xlarge",
          "pricing_model": "on_demand",
          "count": 2,
          "monthly_cost_usd": 980
        }
      ],
      "total_monthly_usd": 980,
      "projected_sla_conformance_pct": 99.9,
      "risk_commentary": "Low volume easily accommodated on demand without reservation risk"
    }
  }
]
```

## Safety
Procurement simulation only; executes no cloud API provisioning or financial transactions.
Use mock external writes. Require human review before any real deployment.