Skip to main content
Back to challenges
Challenge

Build an AI Endpoint Telemetry Realization Auditor with Pydantic AI

Develop an endpoint telemetry processing service to solve the AI execution gap in enterprise fleets. Using Pydantic AI, your agent will process GPU usage metrics from edge devices, validate value realization KPIs, and isolate underutilized AI assets with structured precision.

Workflow AutomationHosted by Vera
Sign in free to start
Challenge brief

What you are building

The core problem, expected build, and operating context for this challenge.

Construct a Pydantic AI pipeline to parse hardware telemetry streams and audit value realization across enterprise AI endpoints.

Delivery guide

How work is evaluated

Evaluation

Evaluate audit classification precision for identifying underutilized AI endpoints.

Datasets

Challenge datasets

Public datasets, plus the private datasets you have access to.

Evaluation rubric

How submissions are scored

These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.

Dimensions
2 scoring checks
Binary
2 pass or fail dimensions
Ordinal
0 scaled dimensions
Dimension 1

gap_flag_accuracy

Correctly flags low utilization endpoints (<5% active)

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Dimension 2

audit_classification_f1

F1 score for underutilization flags • target: 0.94 • range: 0-1

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Learning goals

What you should walk away with

  • Implement Pydantic AI models for enterprise hardware telemetry parsing

  • Calculate ROI metrics by comparing GPU resource allocation against inference yields

  • Generate automated remediation schedules for underutilized local AI models

  • Ensure robust error-handling for malformed telemetry log inputs

Resources and assets

Reference links and supporting material

Dataset notes

Hardware metrics dataset capturing GPU load, host power draw, and model request throughput across corporate endpoints.

Challenge setup

Evaluate audit classification precision for identifying underutilized AI endpoints.

Challenge input

JSON with endpoint_id, active_hours, gpu_utilization_pct, inferences_served

Pydantic AI

Type-safe structured AI agent framework.

Evaluated output

JSON with endpoint_id, execution_gap_flag (bool), efficiency_score

Evaluation checks
  • Correctly flags low utilization endpoints (<5% active)
  • F1 score for underutilization flags • target: 0.94 • range: 0-1
Configured evaluation evidence
  • Benchmark: FinOps_AI_Audit_v1
  • Audit Classification F1 target: 0.94
  • 1 public reference case
Configured runtime
  • 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

Action Space
  • Pydantic AI · Required
  • Zed · Optional
  • Evidently AI · Optional
Orchestration
  • Pydantic AI · Required

Evaluation contract

  • gap_flag_accuracy · Weight 1
  • audit_classification_f1 · Weight 1

Metrics

  • audit_classification_f1 · Target 0.94

Benchmarks

  • FinOps_AI_Audit_v1
  • general

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON with endpoint_id, active_hours, gpu_utilization_pct, inferences_served

    Output: JSON with endpoint_id, execution_gap_flag (bool), efficiency_score

    Execution harness: Python

Evaluation tools

  • Pydantic AI · Agent Framework

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Build an AI Endpoint Telemetry Realization Auditor with Pydantic AI