Skip to content

Classic ML system design: recsys, search ranking, feature pipelines

  • System design
  • Asked often
  • About 2 h 30 min to learn
  • 3 vetted resources
Build a plan that includes this

What this topic covers

Before LLMs, the ML system design round meant designing a recommender, a search ranker, an ads click predictor or a fraud model end to end. The expected shape is a repeatable walk: frame the problem and pick offline and online metrics, sketch candidate generation and ranking stages, decide which features are precomputed and which are computed at request time, pick a baseline model, plan serving and A/B testing, then monitoring and retraining. Applied LLM roles still get this round at many companies, and the same vocabulary (two-tower retrieval, learning to rank, feature stores, training-serving skew, online experiments) shows up inside GenAI designs too. Interviewers look for structure, for asking clarifying questions about scale and latency, and for knowing the real trade-offs rather than naming the newest model.

Leads to

You’re ready when you can

What interviewers check, from the resources below.

  • Run the nine steps on “design a video recommender” in under ten minutes without notes
  • Name one offline metric and one online metric for a ranking system and say why they can disagree
  • Explain candidate generation vs ranking and why two stages beat one
  • Classify features as precomputed vs real-time and describe where each is stored
  • Describe how you would A/B test and then monitor a deployed model
  • List the clarifying questions you ask in project setup (goal, users, latency, personalization, constraints)
  • Explain why you compare against random, heuristic and human baselines before training anything
  • Describe three serving decisions (retraining cadence, confidence thresholds, on-device vs server) and their trade-offs
  • Give one assumption a trained model makes about production data and how you would check it
  • Retell one real recommender design (candidate generation, ranking model, features, metric) in two minutes

What to study, in order

Each link opened and checked by hand; the date says when.

  1. Start here

    Machine Learning System Design (AIMLInterviews) (opens in a new tab)

    alirezadir/AIMLInterviews (GitHub) · Alireza Dirafzoon · 60 min · Free · checked 21 Oct 2026

    Why
    The most-used interview template for the ML design round: a numbered walk from problem framing to monitoring that you can run on any recsys or ranking question.
    What to read
    Read “1. The 9-Step ML System Design Formula (Template)” from “1. Problem Formulation” through “9. Scaling, Monitoring, and Updates”, then under “3. ML System Design Topics” read “Recommendation Systems” and “Search and Ranking (Ads, newsfeed, etc)”.
    What to skip
    The NLP, CV, AV and graph topic lists, “4. ML at big tech companies”, and the GenAI / LLM section (covered by genai-system-design).

  2. Free alternative

    Machine Learning Systems Design: Design a machine learning system (opens in a new tab)

    huyenchip.com · Chip Huyen · 45 min · Free · checked 21 Oct 2026

    Why
    A short, opinionated version of the same round from an author interviewers respect; strongest on the questions to ask before you pick a model and on serving decisions.
    What to read
    Read “Project setup”, “Data pipeline”, “Modeling” (including “Model selection”) and “Serving”, plus the two closing notes.
    What to skip
    The “Training” subsections on debugging, hyperparameter tuning and scaling, which are about training practice rather than system design.

  3. Go deeper

    ML and LLM system design: 800 case studies to learn from (opens in a new tab)

    Evidently AI · Evidently AI · 60 min · Free · checked 21 Oct 2026

    Why
    Interviewers reward candidates who can cite how a real company built the thing; this is a filterable index of production write-ups from company engineering blogs.
    What to read
    In the “Real-world ML and LLM systems” table, filter by the recommender systems tag and the search and ranking tag; pick one recsys and one search ranking write-up and read each end to end, noting features, model stages and metrics.
    What to skip
    The GenAI, CV and NLP tags for this topic; “Get started with Evidently” is a product pitch.

Companies that test System design most

Put Classic ML system design in your plan

Tell us your interview date and hours a day; it lands on the right day, in the right order.

Build my plan