Skip to main content
Back to challenges
Challenge

Real-Time Cyber Breach Loss Severity Estimator using AI SDK and RunPod Hosted Models

Large-scale data breach settlements require rapid financial loss estimation across hundreds of millions of affected user records. You will build a streaming loss estimation microservice using the Vercel AI SDK connected to high-throughput open LLMs deployed on RunPod serverless GPUs. The microservice must achieve under 500ms latency per batch while accurately calculating breach tier payouts.

Machine LearningHosted by Vera
Sign in free to start
Challenge brief

What you are building

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

Develop a high-throughput cyber loss severity estimator using Vercel AI SDK streaming connected to open LLMs deployed on RunPod serverless inference.

Delivery guide

How work is evaluated

Evaluation

Evaluate batch throughput and financial loss calculation precision against historical breach settlement classes.

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

latency_check

Response time must be below 1000ms for batch execution.

Binary check

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

Dimension 2

estimation_mae

Mean Absolute Error relative to standard $117.5M breach settlement scale formula. • target: 1000000 • range: 0-5000000

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

  • Construct a Vercel AI SDK dynamic route using `streamText` and custom tool calls.

  • Deploy Llama-3.1 model endpoints on RunPod Serverless with vLLM acceleration.

  • Implement payout tier calculation algorithms for complex class-action breach settlements.

  • Achieve streaming token throughput exceeds 80 tokens/sec with sub-500ms initial response time.

Resources and assets

Reference links and supporting material

Dataset notes

Synthetic dataset of 200 corporate cyber breach incidents detailing record loss metrics, PII flags, and settlement outcomes.

Challenge setup

Evaluate batch throughput and financial loss calculation precision against historical breach settlement classes.

Challenge input

JSON with breach_id, record_count, pii_types_exposed, credit_monitoring_claimed, jurisdiction

AI SDK

Provides type-safe streaming and tool-calling interfaces for TypeScript backends.

RunPod

Delivers scalable, low-latency vLLM inference hosting for custom open models.

Evaluated output

JSON containing total_estimated_payout, average_payout_per_claimant, risk_tier, and confidence_interval

Evaluation checks
  • Response time must be below 1000ms for batch execution.
  • Mean Absolute Error relative to standard $117.5M breach settlement scale formula. • target: 1000000 • range...
Configured evaluation evidence
  • Benchmark: CyberLossBench-2026
  • Estimation Mae target: 1000000
  • 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
  • RunPod · Required
  • Vercel AI SDK · Optional
Policy Serving
  • Sonar · Optional
Orchestration
  • Vercel AI SDK · Optional

Evaluation contract

  • latency_check · Weight 1
  • estimation_mae · Weight 1

Metrics

  • estimation_mae · Target 1000000

Benchmarks

  • CyberLossBench-2026
  • security

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON with breach_id, record_count, pii_types_exposed, credit_monitoring_claimed, jurisdiction

    Output: JSON containing total_estimated_payout, average_payout_per_claimant, risk_tier, and confidence_interval

    Execution harness: JavaScript

Evaluation tools

  • AI SDK · TypeScript Agent Orchestration
  • RunPod · Serverless GPU Infrastructure

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about Real-Time Cyber Breach Loss Severity Estimator using AI SDK and RunPod Hosted Models