Skip to main content
Loading navigation
Back to challenges

High-Throughput Trade Clearing Anomaly Detector with Vercel AI SDK and Llama 3.3 70B

Modern post-trade clearing platforms process high-frequency equity trades demanding instant anomaly detection for market manipulation. Build a real-time streaming trade monitor using the Vercel AI SDK and Llama 3.3 70B via hosted APIs to analyze trade streams and detect spoofing or layer patterns. Process 500 trade events per second with an F1 score exceeding 0.92.

CybersecurityHosted by Vera
Sign in free to start

What you are building

Build a streaming real-time trade monitoring system using Vercel AI SDK and Llama 3.3 70B to flag trade manipulation in high-frequency equity pipelines.

How work is evaluated

Evaluation

Measures trade monitoring stream throughput and precision in identifying market manipulation events.

Datasets

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

How submissions are scored

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

wash_trade_detection_test

Checks if synthetic wash trade batch is successfully detected

Binary check

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

Dimension 2

f1_score

F1 score on trade anomaly benchmark dataset • target: 0.92 • 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

  • Implement a low-latency trade event consumer using Node.js / TypeScript and Vercel AI SDK

  • Integrate Llama 3.3 70B via Fireworks or Together AI provider endpoints

  • Stream real-time compliance assessments using Vercel AI SDK `streamText` and `generateObject`

  • Detect microsecond pattern anomalies such as quote stuffing and wash trading

Resources

Dataset notes

Streaming trade batch dataset containing 20,000 order book events with interspersed wash trade and quote-stuffing anomalies.

Challenge setup

Measures trade monitoring stream throughput and precision in identifying market manipulation events.

Challenge input

JSON stream of trade records (symbol, price, volume, side, timestamp_ms)

AI SDK

Vercel SDK for streaming output and object parsing

Llama 3.3 70B

High-capacity open-weight intelligence model

Evaluated output

JSON output with flagged_anomaly (boolean), pattern_type, and probability

Evaluation checks
  • Checks if synthetic wash trade batch is successfully detected
  • F1 score on trade anomaly benchmark dataset • target: 0.92 • range: 0-1
Configured evaluation evidence
  • Benchmark: TradeSurv-Eval
  • F1 Score target: 0.92
  • 1 public reference case
Configured runtime
  • JavaScript execution harness
  • javascript sandbox (unavailable on Versalist)

Shows the first task, up to three tools, and up to three evidence items.

View technical recipe

Configured tools

Action Space
  • Vercel AI SDK · Required
  • Llama 3.3 70B · Optional
  • Llama Index · Optional
Orchestration
  • Vercel AI SDK · Required

Evaluation contract

  • wash_trade_detection_test · Weight 1
  • f1_score · Weight 1

Metrics

  • f1_score · Target 0.92

Benchmarks

  • TradeSurv-Eval
  • security

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON stream of trade records (symbol, price, volume, side, timestamp_ms)

    Output: JSON output with flagged_anomaly (boolean), pattern_type, and probability

    Execution harness: JavaScript

Evaluation tools

  • AI SDK · Streaming Agent SDK
  • Llama 3.3 70B · LLM Engine

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about High-Throughput Trade Clearing Anomaly Detector with Vercel AI SDK and Llama 3.3 70B