Skip to main content
Back to challenges
Challenge

Infrastructure Noise Evaluator for Agentic Workloads with LlamaIndex

Build an infrastructure noise evaluator using LlamaIndex to isolate hardware variance from model capability changes in agentic coding benchmarks. Successfully filter out host contention noise to achieve a repeatable variance score below 0.5% across sequential eval runs.

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.

Construct a LlamaIndex RAG and evaluation workflow to identify hardware noise in agent execution traces.

Delivery guide

How work is evaluated

Evaluation

Evaluates the variance coefficient of benchmark scores across repeated runs before and after noise filtering.

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

Repeatability Target Met

Filtered variance coefficient must be lower than 0.005 (0.5%).

Binary check

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

Dimension 2

Variance Reduction Percent

Percentage reduction in evaluation score variance after filtering. • target: 0.93 • range: 0.8-0.99

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 trace analysis index with LlamaIndex for benchmark evaluation.

  • Detect performance anomalies caused by noisy neighbors or CPU/GPU throttling.

  • Filter out infrastructure-induced evaluation failures from model capability failures.

  • Achieve statistical repeatability across repeated agent execution runs.

Resources and assets

Reference links and supporting material

Dataset notes

Dataset of 100 benchmark execution traces with added host hardware contention markers.

Challenge setup

Evaluates the variance coefficient of benchmark scores across repeated runs before and after noise filtering.

Challenge input

JSON array of execution trace logs with system metrics

LlamaIndex

Data framework used to index and query execution traces and system metrics

Evaluated output

JSON report containing raw_variance, filtered_variance, and filtered_samples

Evaluation checks
  • Filtered variance coefficient must be lower than 0.005 (0.5%).
  • Percentage reduction in evaluation score variance after filtering. • target: 0.93 • range: 0.8-0.99
Configured evaluation evidence
  • Benchmark: Agentic Evaluation Reliability Index
  • Variance Reduction Percent target: 0.93
  • 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
  • Llama Index · Required
  • LlamaIndex · Optional
  • Llama 3.2 · Optional

Evaluation contract

  • Repeatability Target Met · Weight 1
  • Variance Reduction Percent · Weight 1

Metrics

  • Variance Reduction Percent · Target 0.93

Benchmarks

  • Agentic Evaluation Reliability Index
  • general

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON array of execution trace logs with system metrics

    Output: JSON report containing raw_variance, filtered_variance, and filtered_samples

    Execution harness: Python

Evaluation tools

  • LlamaIndex · Agent Framework

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Infrastructure Noise Evaluator for Agentic Workloads with LlamaIndex