Build the application

## Build
Process multi-sensor datacenter temperature logs to reconstruct localized thermal runaway events. Hosted Claude analyzes hotspot trajectories and drafts cold-aisle containment action items.

## User
Datacenter Facilities Manager

## Workflow
1. Extract time-series temperature measurements from rack intake, exhaust, and cold-aisle floor tile sensors.
2. Detect spatial heat gradient anomalies indicating air recirculation or perforated floor tile blockages.
3. Submit thermal incident histories to hosted Claude to generate an engineering post-mortem draft.
4. Output a remediation plan detailing rack blanking panel fixes, fan curve modifications, and workload re-balancing.

## Interface
Multi-tier browser interface with rack temperature heatmaps and downloadable incident summaries.

## Data
Construct synthetic CSV files modeling rack row temperatures, perforated tile CFM flows, and incident timestamps.
Starter examples below are synthetic. They are not downloaded records.

## Source
[NVIDIA DCGM exporter](https://raw.githubusercontent.com/NVIDIA/dcgm-exporter/main/README.md)
Leverages DCGM exporter telemetry concepts for multi-GPU thermal monitoring and rack-level thermal tracking.

## Tools
mastra-ai, anthropic, pydantic, pandas.
Mastra AI orchestrates deterministic workflow steps and tool executions for building facility incident reporting agents.
Hosted Claude API for contextual thermal pattern interpretation and engineering post-mortem generation.

## 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
- Isolates the specific server rack and U-position exhibiting the initial thermal gradient breach.
- Generates post-mortem text identifying recirculation vs lack of chilled airflow causes.
- Specifies concrete physical actions including blanking panel inspections and CFM setpoint revisions.

## Starter Examples
```json
[
  {
    "input": {
      "row_id": "Row-G",
      "sensor_records": [
        {
          "rack_id": "Rack-01",
          "u_position": 24,
          "intake_temp_c": 22.5,
          "exhaust_temp_c": 38
        },
        {
          "rack_id": "Rack-02",
          "u_position": 24,
          "intake_temp_c": 34.8,
          "exhaust_temp_c": 54.2
        }
      ]
    },
    "expected_output": {
      "hotspot_identified": true,
      "critical_location": {
        "rack_id": "Rack-02",
        "u_position": 24
      },
      "probable_cause": "AIR_RECIRCULATION_LEAK",
      "remediation_summary": "INSTALL_BLANKING_PANELS_U24"
    }
  },
  {
    "input": {
      "row_id": "Row-H",
      "sensor_records": [
        {
          "rack_id": "Rack-01",
          "u_position": 12,
          "intake_temp_c": 21,
          "exhaust_temp_c": 34
        },
        {
          "rack_id": "Rack-02",
          "u_position": 12,
          "intake_temp_c": 21.5,
          "exhaust_temp_c": 35
        }
      ]
    },
    "expected_output": {
      "hotspot_identified": false,
      "critical_location": null,
      "probable_cause": "NORMAL_OPERATION",
      "remediation_summary": "NO_ACTION_REQUIRED"
    }
  }
]
```

## Safety
Advisory diagnostics only. No direct actuation of datacenter HVAC, chillers, or CRAH fan speeds.
Use mock external writes. Require human review before any real deployment.