All roadmaps
intermediate
~8 weeks · 4-6 hrs/week· 8 stages

ML Engineer — Interview Ready

The 8-week ramp to walking into interviews confident

You'll answer the questions interviewers actually ask across LLMs, RAG, classical ML, systems, and ethics — with senior signals baked in.

Who this is for

You've done some ML work and have interviews coming up (or want to be ready when they do).

Prerequisites
  • You've trained at least one model.
  • Comfortable with Python and basic ML vocabulary.
North-star project

Complete a full mock interview cycle (fundamentals + system design + behavioral) and pass a real one.

Your progress0 of 8 · 0%

Progress is saved on this device. Sign in to sync it across devices.

  1. 1

    Stage 1: Nail the LLM fundamentals

    Every ML interview in 2026 opens here. Miss these and the rest doesn't matter.

    Deliverable · Score 90%+ on the LLM Fundamentals quiz — Junior + Mid combined.
  2. 2

    Stage 2: Prompting & structured output

    Interviewers probe this because it separates candidates who've built things from ones who've read about them.

    Deliverable · Explain few-shot, CoT, self-consistency, and structured output — out loud — in under 60 seconds each.
  3. 3

    Stage 3: RAG + embeddings

    The most-asked applied topic. If you can't design a RAG system on a whiteboard, this stage is why.

    Deliverable · Whiteboard a RAG system for company docs, end-to-end, from ingest to serving.
  4. 4

    Stage 4: Classical ML you'll still be asked about

    Bias-variance, overfitting, metrics, regularization — the classics stay in the interview loop.

    Deliverable · Explain the bias-variance tradeoff and pick a metric for an imbalanced problem — out loud.
  5. 5

    Stage 5: Deep learning & transformers

    Self-attention, positional encoding, the training loop — the questions that separate mid from senior.

    Deliverable · Derive self-attention on paper. Explain why we scale by sqrt(d_k).
  6. 6

    Stage 6: Agents, fine-tuning, evaluation

    The 2026 additions to every ML interview. Show you've kept up.

    Deliverable · Explain when NOT to use an agent, LoRA vs full fine-tuning, and how to measure hallucination — cleanly.
  7. 7

    Stage 7: ML system design + MLOps

    This is where offers are won. Design questions test judgment more than knowledge.

    Deliverable · Design a RAG chatbot and a semantic-search system on a whiteboard, with tradeoffs stated.
  8. 8

    Stage 8: Ethics, safety, and the human side

    Increasingly asked — and a great place to sound senior.

    Deliverable · Ace one full mock interview. Then take a real one.