Build the application

## Build
Analyze multi-accelerator server chassis thermal logs and cooling sensor feeds to detect thermal throttling events. Formulate rack airflow rebalancing plans and workload evacuation proposals.

## User
Data Center Facilities and Compute Engineer

## Workflow
1. Ingest time-series DCGM thermal readings and clock frequency records from physical accelerator nodes.
2. Identify sudden clock speed drops correlated with thermal limit breaches.
3. Map localized hot spots across server chassis fan zones and compute trays.
4. Invoke hosted Claude API to draft a mechanical thermal mitigation and load-shedding plan.

## Interface
Chassis thermal heat-map dashboard with cooling fan status indicators and maintenance export.

## Data
Generate synthetic CSV files with 24-hour sensor telemetry representing 16 GPUs across two rack chassis.
Starter examples below are synthetic. They are not downloaded records.

## Source
[NVIDIA DCGM exporter](https://raw.githubusercontent.com/NVIDIA/dcgm-exporter/main/README.md)
Reference DCGM exporter metrics for accelerator die temperature, slowdown throttling flags, and fan speeds.

## Tools
pydantic-ai, anthropic, matplotlib.
pydantic-ai validates thermal reading schemas and bounds before initiating automated risk assessment.
Hosted Claude API accessed via SDK for root-cause synthesis; no internal hardware measurements asserted.

## 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
- Identifies GPU thermal throttling episodes where temperature exceeds 84C and clocks drop by over 20 percent.
- Highlights chassis fan zone degradation where fan duty cycle is pinned at maximum with rising temps.
- Synthesizes an incident ticket specifying failed fan modules and recommended job rescheduling.

## Starter Examples
```json
[
  {
    "input": {
      "chassis_id": "chassis-03",
      "gpu_index": 6,
      "temperature_c": 89,
      "throttle_flag_active": true,
      "clock_mhz": 1120,
      "nominal_clock_mhz": 1800,
      "fan_speed_pct": 100
    },
    "expected_output": {
      "thermal_state": "CRITICAL_THROTTLE",
      "clock_deficit_pct": 37.78,
      "chassis_risk": "local_airflow_restriction",
      "remedial_action": "evacuate_workload_and_inspect_chassis_fan_b"
    }
  },
  {
    "input": {
      "chassis_id": "chassis-01",
      "gpu_index": 2,
      "temperature_c": 64,
      "throttle_flag_active": false,
      "clock_mhz": 1785,
      "nominal_clock_mhz": 1800,
      "fan_speed_pct": 55
    },
    "expected_output": {
      "thermal_state": "NOMINAL",
      "clock_deficit_pct": 0.83,
      "chassis_risk": "none",
      "remedial_action": "none"
    }
  }
]
```

## Safety
Generates operational cooling recommendations without directly controlling data center building management systems or fan speeds.
Use mock external writes. Require human review before any real deployment.