Skip to main content
Back to challenges
Challenge

Pydantic AI & Claude 4 Sonnet Telemetry & Token Power Attribution Engine

Create a structured telemetry and power attribution engine using Pydantic AI and Claude 4 Sonnet to map token generation to hardware energy metrics. Build energy-aware cost calculators that accurately attribute watt-hours and cloud expenses per tenant with under 2% error.

EngineeringHosted by Vera
Sign in free to start
Challenge brief

What you are building

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

Design a type-safe power and cost telemetry attribution system using Pydantic AI and Claude 4 Sonnet to model hardware energy per output token.

Delivery guide

How work is evaluated

Evaluation

Evaluates accuracy of per-tenant power and energy attribution metrics.

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

error_below_threshold

Ensures attribution error is under 2.0%.

Binary check

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

Dimension 2

joules_per_token_accuracy

Precision of calculated energy per token. • target: 0.99 • 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

  • Build a Pydantic AI agent with structured validation for dynamic hardware telemetry parsing

  • Integrate OpenTelemetry metrics and NVIDIA Management Library (NVML) power usage traces

  • Attribute real-time GPU/TPU wattage and energy usage (Joules/token) to distinct multi-tenant API users

  • Achieve less than 2% attribution discrepancy against hardware energy meters

Resources and assets

Reference links and supporting material

Dataset notes

Dataset of GPU power draw logs recorded at 100ms intervals during multi-tenant parallel LLM generation sessions.

Challenge setup

Evaluates accuracy of per-tenant power and energy attribution metrics.

Challenge input

JSON telemetry log with timestamped token outputs and power trace samples (Watts).

Pydantic AI

Type-safe Python agent framework with strict validation.

Claude 4 Sonnet

Powers deep telemetry reasoning and report generation.

Evaluated output

JSON breakdown of total Joules, Joules/token, and estimated electricity cost per tenant.

Evaluation checks
  • Ensures attribution error is under 2.0%.
  • Precision of calculated energy per token. • target: 0.99 • range: 0-1
Configured evaluation evidence
  • Benchmark: Green AI Hardware Telemetry Standard
  • Joules Per Token Accuracy target: 0.99
  • 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
  • Claude Code · Optional
Observation
  • Claude Code · Optional
Policy Serving
  • Claude 4 Sonnet · Optional
Orchestration
  • Pydantic AI · Required

Evaluation contract

  • error_below_threshold · Weight 1
  • joules_per_token_accuracy · Weight 1

Metrics

  • joules_per_token_accuracy · Target 0.99

Benchmarks

  • Green AI Hardware Telemetry Standard
  • energy

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON telemetry log with timestamped token outputs and power trace samples (Watts).

    Output: JSON breakdown of total Joules, Joules/token, and estimated electricity cost per tenant.

    Execution harness: Python

Evaluation tools

  • Pydantic AI · Agent Framework
  • Claude 4 Sonnet · Models

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Pydantic AI & Claude 4 Sonnet Telemetry & Token Power Attribution Engine