Interview Prep — Roadmap

Once you've covered the topic sections, this is where you drill for the actual interview loop. Structured after the modules used by FAANG-style AI/ML interviews.

1. General Coding (DSA)

• Arrays, strings, hash maps, two pointers, sliding window
• Trees, graphs, BFS/DFS
• Dynamic programming
• Practice on LeetCode-style problems, timed

2. ML Coding

• Implement classical algorithms from scratch (k-means, k-NN, logistic regression, decision tree)
• Implement a neural network forward/backward pass without a framework
• Implement attention from scratch

3. ML/GenAI Knowledge Q&A

• Classical ML breadth questions (bias-variance, regularization, evaluation metrics)
• Deep learning breadth questions (architectures, training dynamics)
• LLM/GenAI breadth questions (transformers, RAG, fine-tuning, agents)
• Inference engineering breadth questions (KV cache, batching, quantization, TTFT/TPOT)
• Research-track questions (paper critique, ablation design, literature fluency)
• Rapid-fire self-check: can you explain each topic in under 2 minutes, out loud?

3.5. Technology Comparisons & Decision Trees

• vLLM vs SGLang vs TensorRT-LLM vs Triton vs llama.cpp
• LangChain vs LangGraph
• RAG vs fine-tuning
• LoRA vs QLoRA vs full fine-tuning
• SQL vs vector DB vs graph DB
• DDP vs FSDP vs ZeRO
• CUDA vs ROCm
• PyTorch vs JAX

4. System Design

• Practice the 9-step ML system design formula on 5-10 different problems (see ML System Design roadmap)
• Practice LLM/RAG/agent system design specifically — increasingly common in 2026-era interviews
• Time yourself: most system design rounds are 45 minutes

5. Behavioral

• Prepare stories using the STAR method (Situation, Task, Action, Result)
• Have examples ready for: conflict with a teammate, a failed project, a time you influenced without authority
• Research the specific company/team's ML maturity and tailor answers