Mortgage Bond Prepayment Risk Forecasting Stream with AI SDK & Mastra AI
AI-driven refinancing platforms accelerate mortgage prepayments, destabilizing traditional MBS yield models. Build an interactive AI SDK dynamic agent stream backed by Mastra AI workflow nodes to forecast prepayment velocity and portfolio interest rate sensitivity within sub-200ms latency limits.
What you are building
The core problem, expected build, and operating context for this challenge.
Implement a streaming financial risk analysis engine using Vercel AI SDK data streaming and Mastra AI workflow components.
How work is evaluated
Evaluates accuracy of dynamic CPR prepayment velocity and yield calculations in real-time response streams.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
cpr_range_test
Ensures CPR calculation stays within expected quantitative limits
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
cpr_mean_absolute_error
MAE against theoretical prepayment benchmark model • target: 0.5 • range: 0-5
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Stream dynamic MBS portfolio risk recalculations using Vercel AI SDK AI Stream response handlers.
Execute prepayment vector evaluations through embedded Mastra AI data nodes.
Calculate Conditional Prepayment Rates (CPR) under changing interest rate shocks.
Maintain real-time streaming latency under 200 milliseconds.
Reference links and supporting material
Historical and synthetic MBS pool data, current interest rate curves, and simulated AI refinancing adoptability ratios.
Challenge setup
Evaluates accuracy of dynamic CPR prepayment velocity and yield calculations in real-time response streams.
Challenge input
JSON containing pool_wac, current_rate, ai_refi_penetration
AI SDK
Framework assignment for streaming multi-provider dynamic agent UI integration.
Mastra AI
Additional tool for complex workflow step coordination.
Evaluated output
JSON with projected_cpr, price_duration_impact
- Ensures CPR calculation stays within expected quantitative limits
- MAE against theoretical prepayment benchmark model • target: 0.5 • range: 0-5
- Benchmark: FinRisk-Stream-Bench
- Cpr Mean Absolute Error target: 0.5
- 1 public reference case
- 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
- Mastra AI · Required
- Hamming · Optional
- Arize Phoenix · Optional
- Arize Phoenix · Optional
- Mastra AI · Required
Evaluation contract
- cpr_range_test · Weight 1
- cpr_mean_absolute_error · Weight 1
Metrics
- cpr_mean_absolute_error · Target 0.5
Benchmarks
- FinRisk-Stream-Bench
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing pool_wac, current_rate, ai_refi_penetration
Output: JSON with projected_cpr, price_duration_impact
Execution harness: JavaScript
Evaluation tools
- AI SDK · Streaming Agent Framework
- Mastra AI · Agent Systems Framework
Recipe freeze marker
No recipe freeze time is recorded.