AI-Assisted Loan Decisioning Engine
Northwind Analytics
Northwind's credit team was stuck between a brittle rules engine and a growing backlog of edge cases. Auditors wanted a paper trail; analysts wanted a model that could adapt. We built a hybrid pipeline that scores risk, explains itself, and cites the policy it used.
01 — The brief
The challenge
Northwind's underwriting team relied on a rigid rules engine that couldn't adapt to new risk patterns, while manual reviews created a decision backlog. Auditors also struggled to get consistent, defensible explanations for edge-case decisions.
02 — What we built
The solution
We built a hybrid pipeline combining the existing rules engine with a PyTorch risk model, deployed on SageMaker, plus a retrieval-augmented assistant that lets analysts query underlying policy documents to generate audit-ready explanations.
03 — Method
How we got there
- 01Trained and validated a risk-scoring model against 5 years of historical outcomes
- 02Deployed the model on AWS SageMaker with automated retraining pipelines
- 03Built a RAG assistant over internal policy documentation for analysts
- 04Established a full audit trail linking each decision to its contributing factors
04 — Proof
What changed
- 43% faster average decision turnaround
- 99.2% model uptime over 12 months
- Full audit-trail compliance achieved
“The RAG-based decisioning assistant they built has become the single most-used internal tool at our company. Auditors love the transparency.”
05 — Frames
Inside the work
Ready to build something great?
Tell us about your project and we'll get back to you within one business day with next steps.