ML Design Fundamentals

ML System Design Interview Wiki

Bridge the chasm between research models and production systems. Deconstruct end-to-end architectures for recommendation engines and real-time inference, mastering the critical trade-offs in data freshness, latency, and scalability expected at the world's leading AI labs.

ML Design

Two-Tower Model

Definition

A dual-encoder architecture consisting of two separate neural networks (towers) that map queries (users) and candidates (items) into a shared d-dimensional embedding space, where similarity is calculated via a simple dot product or cosine distance.

ML Design

MMoE

Definition

Multi-gate Mixture-of-Experts (MMoE) is a Multi-Task Learning (MTL) architecture that adapts the Mixture-of-Experts structure by using task-specific gating networks to weight shared expert sub-networks, allowing the model to learn complex task relationships and mitigate negative transfer.

ML Design

Collaborative Filtering

Definition

A recommendation strategy that predicts a user's interests by collecting preferences from many users, leveraging the 'wisdom of the crowd' rather than item metadata.

ML Design

MMR Re-ranking

Definition

Click to auto-generate concept

ML Design

FAISS/HNSW

Definition

Click to auto-generate concept

ML Design

DeepFM

Definition

DeepFM is a hybrid deep learning model that integrates Factorization Machines (FM) and Deep Neural Networks (DNN) to model low-order and high-order feature interactions simultaneously, utilizing a shared embedding layer for both components.

ML Design

XGBoost/LightGBM

Definition

Gradient Boosted Decision Trees (GBDT) implementations that optimize for speed and performance through second-order Taylor expansion and efficient split-finding algorithms.

ML Design

DCN

Definition

The Deep & Cross Network (DCN) is a neural network architecture designed to learn explicit, bounded-degree feature interactions efficiently alongside deep non-linear representations, specifically for large-scale categorical and sparse data.

ML Design

Logistic Regression

Definition

A probabilistic linear model that estimates the probability of a binary outcome by passing a linear combination of features through the logistic (sigmoid) function.

ML Design

Embeddings

Definition

Click to auto-generate concept

ML Design

LSTM

Definition

A specialized Recurrent Neural Network (RNN) architecture designed to model long-range dependencies in sequential data by utilizing a gated memory cell to mitigate the vanishing gradient problem.

ML Design

Isolation Forest

Definition

Click to auto-generate concept

The Definitive Standard

The Foundational Knowledge for Real Interviews.

Every concept explained with the depth, trade-offs, and failure modes that differentiate L5 from L6 engineers at Big Tech.

First-Principles Explanations

Real Interview Trade-offs

Six Technical Domains

Continuously Updated

12+ Concepts

Engineered using the same evaluation criteria employed by senior leadership at Google, Meta, and OpenAI.

20+ Concepts

System Design

Master the blueprints of planet-scale infrastructure. From distributed consensus to eventual consistency, learn the architectural patterns that distinguish Staff Engineers and develop the intuition to navigate complex trade-offs where there are no right answers—only better choices.

Algorithms

Coding Puzzle

Elevate your problem-solving from pattern matching to algorithmic mastery. Build a deep mental library of core data structures and advanced techniques, enabling you to decompose complex problems, optimize with surgical precision, and write clean, industrial-grade code under intense scrutiny.

AI Specialization

ML System Design

Leadership

Behavioral & Leadership

Command the room with high-stakes leadership narratives. Learn to deconstruct your professional journey through the lens of executive-level competencies—conflict resolution, strategic ownership, and technical influence—to demonstrate the maturity and impact required for senior leadership roles.

Data Mastery

SQL Puzzle

Master the engine behind the data. Go beyond basic queries to understand the mechanics of indexing, query planning, and relational theory, empowering you to design bulletproof schemas and optimize data pipelines that handle massive throughput with uncompromising reliability.

Architecture

Frontend System Design

Architect modern web applications that scale with complexity. Dive into the nuances of state management, rendering strategies, and performance optimization, learning how to build resilient frontend systems that deliver premium user experiences across diverse environments and network conditions.

Your Complete Tech Interview Reference

The Knowledge Hub is designed to be your go-to reference — the kind of resource you bookmark and return to again and again. Whether you're brushing up on Kafka's pub-sub internals, revisiting consistent hashing before a system design round, or looking up dynamic programming patterns for a coding interview, every concept is explained with the depth and precision that top-tier interviews demand.

Each article focuses on the why behind a technology — the trade-offs, real-world failure modes, and interview angles that interviewers at Google, Meta, Amazon, and Apple actually probe. This is not a surface-level overview site. It's a reference engineered for engineers.

What Makes a Great Technical Reference?

The best references are flat, fast, and searchable — not structured into rigid beginner-to-advanced hierarchies. System design, ML systems, and algorithms are sibling disciplines, not a linear ladder. You can tackle Kafka and transformers in the same afternoon. Use the search bar and category filters to navigate exactly where you need to go, when you need it.

Interview-Focused Depth

Interviewers rarely ask “define Kafka.” They ask you to design a notification system and justify your messaging choice. Every article covers the trade-offs and failure modes that separate L5 from L6 candidates.

Six Technical Domains

System Design, Coding Algorithms, ML Design, Frontend Architecture, SQL & Databases, and Behavioral frameworks — all in one place. Filter to focus, or search across everything at once.

First-Principles Thinking

Memorizing answers fails in follow-up questions. Every concept is grounded in first principles — understand CAP theorem not as a definition, but as a real constraint you'd navigate in a production incident.