Skip to main content
Back to challenges
Challenge

Build GPU Infrastructure Workload Allocator with AI SDK and Milvus

Cloud data centers face GPU cost volatility and resource allocation inefficiencies. Build a TypeScript allocation agent using Vercel AI SDK and Milvus vector search to continuously match deep learning workloads with optimal GPU clusters, reducing compute costs by over 20%.

Business OperationsHosted 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 an intelligent GPU workload scheduler leveraging Vercel AI SDK streaming and Milvus vector similarity search.

Delivery guide

How work is evaluated

Evaluation

Evaluates GPU workload node assignment efficiency and cost reduction score.

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

budget_compliance

Ensures allocated instance cost does not exceed input max budget

Binary check

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

Dimension 2

cost_efficiency

Cost savings metric compared to baseline random allocation • target: 0.25 • 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 dynamic tool calling using Vercel AI SDK (`ai`) in TypeScript

  • Perform high-dimensional workload profile indexing and retrieval using Milvus vector database

  • Optimize GPU server cluster selection for minimum cost and latency

Resources and assets

Reference links and supporting material

Dataset notes

Telemetry logs of GPU node pricing and utilization metrics across cloud regions.

Challenge setup

Evaluates GPU workload node assignment efficiency and cost reduction score.

Challenge input

JSON containing batch_size, model_params_billions, memory_gb, max_budget_usd

AI SDK

Provides seamless TypeScript LLM streaming and function calling

Milvus

Enables fast vector similarity search over cluster profiles

Evaluated output

JSON containing assigned_instance, cost_per_hour, and vector_distance

Evaluation checks
  • Ensures allocated instance cost does not exceed input max budget
  • Cost savings metric compared to baseline random allocation • target: 0.25 • range: 0-1
Configured evaluation evidence
  • Benchmark: Cloud-GPU-Bench
  • Cost Efficiency target: 0.25
  • 1 public reference case
Configured runtime
  • JavaScript execution harness
  • javascript 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
  • Milvus · Required
  • Vercel AI SDK · Optional
  • Bland AI · Optional
Orchestration
  • Vercel AI SDK · Optional

Evaluation contract

  • budget_compliance · Weight 1
  • cost_efficiency · Weight 1

Metrics

  • cost_efficiency · Target 0.25

Benchmarks

  • Cloud-GPU-Bench
  • general

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON containing batch_size, model_params_billions, memory_gb, max_budget_usd

    Output: JSON containing assigned_instance, cost_per_hour, and vector_distance

    Execution harness: JavaScript

Evaluation tools

  • AI SDK · Agent Framework
  • Milvus · Inference & Model Runtime

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Build GPU Infrastructure Workload Allocator with AI SDK and Milvus