Bank Insolvency & Liquidity Risk Early Warning System with LangGraph
Small bank closures create systemic contagion risk across regional financial markets when liquidity stress signals are detected too late. Build a stateful bank liquidity monitoring workflow using LangChain and LangGraph to track balance sheet risk vectors, achieving 94% early failure prediction accuracy.
What you are building
Build a stateful, cyclic multi-step agent workflow using LangGraph to monitor liquidity coverage ratios, deposit outflow velocities, and regulatory risk thresholds.
How work is evaluated
Evaluates early warning failure prediction accuracy and state transitions in LangGraph workflows.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
State Checkpointing Test
Verify graph state is correctly stored and restored from memory checkpointer.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
insolvency_prediction_accuracy
Accuracy in identifying troubled institutions 60 days prior to closure • target: 0.94 • range: 0.9-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Design stateful cyclic graph architectures with LangGraph and StateGraph
Implement custom agent state schemas for multi-period financial ratio tracking
Integrate automated FDIC receiver action early warning alerts based on stress triggers
Resources
Synthetic financial report time-series dataset covering 500 small to mid-sized regional bank balance sheets.
Challenge setup
Evaluates early warning failure prediction accuracy and state transitions in LangGraph workflows.
Challenge input
JSON balance sheet time-series data over 4 quarters
LangChain
Assigned framework ecosystem.
LangGraph
Mandated stateful graph workflow library.
Evaluated output
JSON risk evaluation with state summary, distress severity score, and recommended intervention
- Verify graph state is correctly stored and restored from memory checkpointer.
- Accuracy in identifying troubled institutions 60 days prior to closure • target: 0.94 • range: 0.9-1
- Benchmark: BankRisk-Bench
- Insolvency Prediction Accuracy target: 0.94
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Shows the first task, up to three tools, and up to three evidence items.
View technical recipe
Configured tools
- LangGraph · Required
- Langchain · Optional
- LangChain · Optional
- LangGraph · Required
- Langchain · Optional
- LangChain · Optional
Evaluation contract
- State Checkpointing Test · Weight 1
- insolvency_prediction_accuracy · Weight 1
Metrics
- insolvency_prediction_accuracy · Target 0.94
Benchmarks
- BankRisk-Bench
- security
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON balance sheet time-series data over 4 quarters
Output: JSON risk evaluation with state summary, distress severity score, and recommended intervention
Execution harness: Python
Evaluation tools
- LangChain · Agent Framework
- LangGraph · Stateful Orchestration Engine
Recipe freeze marker
No recipe freeze time is recorded.