Deploy an Account Aggregator Credit Underwriting Workflow with LangChain and Daytona
India's Account Aggregator (AA) ecosystem enables consent-driven financial data sharing. In this challenge, you will construct an automated credit underwriting workflow using LangChain stateful agents (LangGraph) hosted inside isolated Daytona developer environments. The system will request financial data artifacts via consent handles, analyze bank account statements, calculate debt-to-income and cash flow metrics, and output a structured credit risk recommendation while verifying digital signatures and consent expiration.
What you are building
The core problem, expected build, and operating context for this challenge.
Build consent-aware credit assessment pipelines using LangChain agents executed in secure Daytona environments.
How work is evaluated
Evaluates accuracy of financial analysis and adherence to consent constraints.
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.
underwriting_decision_test
Checks correct credit decision generation for solvent profile.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
scoring_precision
Accuracy of financial capability evaluation. • target: 0.95 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Design LangChain/LangGraph state machines for consent-based financial data fetching.
Set up isolated Daytona agent workspaces for secure processing of sensitive PII.
Parse and validate Account Aggregator XML/JSON financial statement payloads.
Compute automated creditworthiness, cash balance volatility, and default risk scores.
Reference links and supporting material
Synthetic financial transaction history and consent metadata formatted according to ReBIT AA specifications.
Challenge setup
Evaluates accuracy of financial analysis and adherence to consent constraints.
Challenge input
JSON with consent_handle, applicant_id, monthly_inflow_avg, monthly_outflow_avg, and bounce_count.
LangChain
Required framework for building stateful agent graph workflows.
Daytona
Required environment orchestrator for safe isolated executions.
Evaluated output
JSON with credit_score, approval_status, max_credit_limit, and risk_factors.
- Checks correct credit decision generation for solvent profile.
- Accuracy of financial capability evaluation. • target: 0.95 • range: 0-1
- Benchmark: Financial Risk Scoring Standard
- Scoring Precision target: 0.95
- 1 public reference case
- Python execution harness
- Python 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
- Langchain · Required
- LangChain · Optional
- Larridin · Optional
- Langchain · Required
- LangChain · Optional
Evaluation contract
- underwriting_decision_test · Weight 1
- scoring_precision · Weight 1
Metrics
- scoring_precision · Target 0.95
Benchmarks
- Financial Risk Scoring Standard
- security
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON with consent_handle, applicant_id, monthly_inflow_avg, monthly_outflow_avg, and bounce_count.
Output: JSON with credit_score, approval_status, max_credit_limit, and risk_factors.
Execution harness: Python
Evaluation tools
- LangChain · Agent Framework
- Daytona · Environment Runtime
Recipe freeze marker
No recipe freeze time is recorded.