Stablecoin Settlement Engine with AI SDK and Gemini 2.5 Flash
Cross-border stablecoin transactions require real-time compliance validation and liquidity routing before final ledger settlement. Build an automated settlement agent using Vercel AI SDK and Gemini 2.5 Flash to inspect transaction metadata, run AML rules, and achieve 98% classification accuracy under 300ms latency.
What you are building
Implement a real-time stablecoin compliance and settlement verification pipeline using Vercel AI SDK paired with Gemini 2.5 Flash.
How work is evaluated
Evaluates transaction risk classification accuracy, execution latency, and adherence to structured tool calling schemas.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
schema_conformance
Checks if output strictly matches expected Zod schema structure.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
classification_accuracy
Ratio of correctly classified safe vs risky transactions. • target: 0.98 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
p95_latency_ms
95th percentile latency per transaction evaluation in milliseconds. • target: 300 • range: 0-1000
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Stream transaction payload verification using Vercel AI SDK tools
Integrate Gemini 2.5 Flash for low-latency transaction risk evaluation
Implement structured schema validation for stablecoin mint/burn compliance
Enforce sub-300ms SLA for automated settlement approval workflows
Resources
Synthetic dataset of 1,000 cross-border stablecoin settlement payloads containing wallet histories, geo-location IPs, and transfer amounts.
Challenge setup
Evaluates transaction risk classification accuracy, execution latency, and adherence to structured tool calling schemas.
Challenge input
JSON array of transaction payloads including sender, receiver, amount, currency, and origin IP
AI SDK
TypeScript-native agent orchestration framework.
Gemini 2.5 Flash
Low latency inference model ideal for high-throughput streaming evaluation.
Evaluated output
JSON array of decision objects containing decision, risk_score, and rule_triggers
- Checks if output strictly matches expected Zod schema structure.
- Ratio of correctly classified safe vs risky transactions. • target: 0.98 • range: 0-1
- 95th percentile latency per transaction evaluation in milliseconds. • target: 300 • range: 0-1000
- Benchmark: FintechGuard-AML-v1
- Classification Accuracy target: 0.98
- 1 public reference case
- 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
- Vercel AI SDK · Optional
- Google Gemini · Optional
- Gemini 2.5 Flash · Required
- Vercel AI SDK · Optional
Evaluation contract
- schema_conformance · Weight 1
- classification_accuracy · Weight 1
- p95_latency_ms · Weight 1
Metrics
- classification_accuracy · Target 0.98
- p95_latency_ms · Target 300
Benchmarks
- FintechGuard-AML-v1
- security
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON array of transaction payloads including sender, receiver, amount, currency, and origin IP
Output: JSON array of decision objects containing decision, risk_score, and rule_triggers
Execution harness: JavaScript
Evaluation tools
- AI SDK · Agent Orchestration Framework
- Gemini 2.5 Flash · LLM Inference Engine
Recipe freeze marker
No recipe freeze time is recorded.