The 6-Step Framework
1. Problem Framing (5 min)
- What is the exact business objective?
- What does success look like (KPI)?
- Is this supervised? Online or batch?
2. Data Strategy (5 min)
- What data exists? Volume, freshness, quality?
- How to handle missing labels?
- Privacy/compliance constraints?
3. Feature Engineering (10 min)
- Raw features available
- Computed/aggregated features
- Embedding features (text, images)
4. Modeling (10 min)
- Baseline (rule-based)
- ML approach (gradient boosting / NN)
- Evaluation metric and why
5. Serving (10 min)
- Batch or real-time?
- Latency requirements
- Feature store design
6. Monitoring (5 min)
- Data drift, concept drift
- A/B testing plan
- Retraining trigger