Build the application

## Build
Analyze multi-rack liquid cooling manifold pressure and flow rate telemetry to detect localized coolant blockages. Generate pump speed adjustments and maintenance tasks using hosted Claude.

## User
Datacenter Facilities Engineer

## Workflow
1. Collect coolant pressure drops and liter-per-minute flow rates from rack manifold sensor telemetry.
2. Calculate pressure delta between supply and return lines to identify anomalous flow resistance indicating micro-channel blockages.
3. Invoke hosted Claude to generate prioritized loop flushing and pump speed mitigation recommendations based on the calculated delta.
4. Deliver an HVAC maintenance checklist with highlighted node chassis, pressure deltas, and flow constraints.

## Interface
Diagnostics dashboard displaying rack liquid cooling elevations, pressure deltas, and maintenance task checklists.

## Data
Create synthetic structured sensor files with manifold supply and return pressures, plus flow rates, across 42U rack liquid loops.
Starter examples below are synthetic. They are not downloaded records.

## Source
[NVIDIA DCGM exporter](https://raw.githubusercontent.com/NVIDIA/dcgm-exporter/main/README.md)
Parses custom exposed prometheus metrics mimicking DCGM-style environmental telemetry for liquid flow sensors.

## Tools
anthropic-claude, pydantic-ai, pandas.
anthropic-claude converts telemetry-based hydrodynamic anomalies into actionable physical facility maintenance and flushing tasks.
Hosted Claude API for mechanical liquid cooling diagnostics; local logic for pressure delta calculations.

## 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
- Accurately compute the pressure delta by subtracting return pressure from supply pressure.
- Flag nodes where pressure delta exceeds 25 kPa and flow rate is below 1.0 LPM as experiencing blockages.
- Generate concrete remedial steps including localized loop flushes using Claude.

## Starter Examples
```json
[
  {
    "input": {
      "rack_id": "Row-04-Rack-L",
      "note": "Simplified synthetic app records",
      "sensors": [
        {
          "node": "U40",
          "supply_pressure_kpa": 150,
          "return_pressure_kpa": 120,
          "flow_rate_lpm": 0.8
        },
        {
          "node": "U20",
          "supply_pressure_kpa": 150,
          "return_pressure_kpa": 140,
          "flow_rate_lpm": 1.5
        }
      ]
    },
    "expected_output": {
      "blockage_detected": true,
      "affected_nodes": [
        "U40"
      ],
      "max_pressure_delta_kpa": 30,
      "remedial_actions": [
        "Inspect cold plate micro-channels at U40 for particulate buildup.",
        "Schedule localized loop flush during next maintenance window."
      ]
    }
  },
  {
    "input": {
      "rack_id": "Row-04-Rack-M",
      "note": "Simplified synthetic app records",
      "sensors": [
        {
          "node": "U10",
          "supply_pressure_kpa": 150,
          "return_pressure_kpa": 142,
          "flow_rate_lpm": 1.6
        }
      ]
    },
    "expected_output": {
      "blockage_detected": false,
      "affected_nodes": [],
      "max_pressure_delta_kpa": 8,
      "remedial_actions": [
        "Coolant flow resistance is within acceptable baseline parameters."
      ]
    }
  }
]
```

## Safety
Outputs physical maintenance recommendations; operates completely out-of-band from primary facility pumping controllers.
Use mock external writes. Require human review before any real deployment.