ML System Design Case Studies¶
Twenty worked ML and GenAI system design case studies, built for interview prep. Each one drives a single problem to the floor through the same reusable scaffold: clarify, frame as ML, data and labels, baseline, model, evaluation, deployment, monitoring. Every case study includes a full one hour interview transcript, a junior versus senior contrast, and a one page whiteboard cheat sheet.
Read one case study end to end, internalize the scaffold, and the rest become the same skeleton with different nouns.
Recommendation & Personalization¶
How systems decide what to show a user, and how feedback loops quietly bias the labels you train on.
- Recommendation System
- News Feed / Personalized Ranking
- Conversational Recommender
- People You May Know / Graph Recommendation
- Autocomplete / Typeahead Personalization
- Notification Optimization / Bandit System
Search, Ranking & Ads¶
Ranking under a relevance label problem, calibration, and the auction and experimentation layer that sits on top.
LLM & GenAI Systems¶
Grounding, retrieval, agents, serving, safety, and how you evaluate systems that have no clean ground truth.
- Enterprise AI Copilot
- Production RAG / Enterprise Search
- LLM Evaluation & Monitoring Platform
- AI Agent for Customer Support
- Multi-Tenant LLM Serving Platform
- LLM Safety Gateway
- Document Intelligence Pipeline
Trust, Safety & Anomaly¶
Imbalanced, adversarial, and contested label problems where the cost of a mistake is asymmetric.
- Fraud / Anomaly Detection System
- Content Moderation / Policy Enforcement
- Spam / Bot Detection System
ML Platform & Ops¶
The infrastructure underneath every model: safe rollout, train serve skew, drift, and retraining.