Model design review — proposing a new approach

Lesson 2 of 4 in Talking About AI at Work — Meetings, Stand-ups & Reviews · 18 minutes

Realistic ML design review where IEP engineer proposes a new model architecture. Team probes assumptions and identifies risks.

What you'll learn

Present ML design proposals confidently
Handle design review pushback professionally
Discuss alternatives considered
Identify and address risks honestly

From the transcript

Sandeep, IEP ML engineer at a Toronto fintech, is proposing a new model architecture for fraud detection. Current model is gradient boosting; he wants to try a transformer-based approach. Design review with senior ML engineer (Maria), data scientist (Priya), and engineering manager (David). 60-90 minute meeting; Sandeep has 15 minutes to present, then Q&A.

Sandeep: Thanks everyone. I want to walk through my proposal for the next-generation fraud model. Goal is to share the rationale, get your pushback, and align on next steps. I'll take 15 minutes; then 30 minutes for questions.

Sandeep: Current state: our gradient boosting model achieves 0.87 AUC on validation. Production performance has…