Agent 04 · Fintech
LoanProcess Agent
Repairs loan applications and flags high-risk ones with a model score and reasons.
Agent PackLendFlowLangGraph
One offline run on demo data with planted problems (recorded platform test, November 2026).
- Client problem
- Incomplete applications stall underwriting; officers re-key data and miss risk signals.
- What it does
- Finds missing employment length and interest rate, impossible ages, then scores default risk with a logistic-regression model.
- How it heals
- Applies the rate card for a missing interest rate, requests employment proof, sends impossible ages for verification, and routes high-risk files to a senior credit officer with the reasons.
- Human gate
- Credit decisions stay with the loan officer.
- Typical buyer
- NBFCs, small lenders (credit head)
- Built with
- LangGraph self-healing loop
- Demo data
- Kaggle laotse/credit-risk-dataset (32,581 loans).
- Tables
- loan_applications ← Kaggle laotse/credit-risk-dataset
- Target outcome
- Target: 30-50% less underwriting prep time.
- Path to production
- Plug into Account Aggregator and bureau pulls; officer approves every decision.
| Issue | Risk | Proposed fix | Count |
|---|---|---|---|
| missing_person_emp_length 'person_emp_length' is empty | medium | Request employment proof from applicant | 8 |
| missing_loan_int_rate 'loan_int_rate' is empty | low | Apply rate card for the loan grade | 8 |
| out_of_range_person_age person_age=144 outside [18, 100] | medium | Send for age / DOB verification | 5 |
| high_default_risk Default probability 100%, loan 9,075 at 187% of income | medium | Route to senior credit officer with risk summary | 20 |
How we demo it
- Load 4,000 credit-risk applications with blanks and age=144 errors.
- Run LoanProcess: fixes applied, risk bands added.
- Filter to high risk and read the reasons.
Built on AWS
Runs in the platform container on EC2 Graviton (or on demand in Lambda), writes healed tables to the S3 lakehouse, uses Amazon Bedrock only to write explanations, and reports six metrics to CloudWatch.
FraudShield
Explains and triages risky card transactions for analysts.
02 · FintechKYC AutoFlow
Checks KYC records, catches bad documents and duplicates before they reach underwriting.
03 · FintechAML Monitor
Watches mobile-money flows for structuring and round-tripping, and drafts case files.
Step into DigiQuench.
Agents that heal your data.
Watch an agent find planted problems, fix the safe ones and wait for your approval on the rest. Thirty minutes, live.