A lending engine that turns a raw application into a priced, funded decision in 90 seconds, with no human in the loop. It moves $8M+ in originations every month at four-nines uptime.
Traditional underwriting takes days because a human reads bank statements by hand. We replaced that with a Plaid pull and a Python scoring service, so the applicant connects their bank and the model answers before they close the tab.
Speed is a conversion lever: every minute of friction drops funnel completion, and an instant priced offer is the difference between a funded loan and an abandoned one.
Automation in lending is only safe if it is defensible. A deterministic rules engine sits on top of the ML score, so regulatory limits and lender policy are enforced in code, not buried in a model.
Every decision is persisted with its full feature lineage in Postgres, letting compliance replay any approval or decline exactly as it ran, months later.
| Component | Technology | Why |
|---|---|---|
| API & Orchestration | Node.js | Async I/O fans out Plaid and scoring calls without blocking. |
| Bank Connectivity | Plaid | One integration reaches 12k+ institutions for cash-flow data. |
| Scoring Engine | Python | Native ML stack runs the underwriting model and feature pipeline. |
| System of Record | PostgreSQL | ACID feature store plus the immutable decision ledger. |
| Infrastructure | AWS | Multi-AZ ECS, SQS, and Lambda power the funding workflow. |
Plaid Link connects an applicant's accounts in-flow without uploading a single statement.
The model scores real transaction history, not just a credit bureau pull, widening approvals.
A rules engine enforces regulatory and lender constraints on top of every ML score.
Approvals trigger an SQS-backed AWS workflow for KYC, disbursement, and notifications.
Each decision is stored with full feature lineage for compliance replay.
Stateless services across availability zones hold 99.9% uptime under load.
“"The applicant connects their bank, and the loan decides itself before they close the tab."”
The platform replaced days of manual review with a 90-second automated decision, cutting cost per decision by 40% while sustaining $8M+ in monthly originations at 99.9% uptime. By pairing a Python ML scoring engine with a deterministic policy layer and an immutable Postgres audit ledger, it delivers the speed of full automation without surrendering the auditability fintech lending demands, fast, defensible, and built to scale.