100% Free · No Login Required · Math, Code, Follow-ups & Review

Free ML & LLM Interview Practice

Each topic connects math derivations, from-scratch code, edge-case bugs, and interview follow-ups. No sign-up required — search or pick a track below.

🔍⌘K

💻ML Kernels & Numerics

105 Drills

Implement core kernels in NumPy with runnable assertions, overflow checks, and backprop notes.

Browse Kernels➔

📐Theory & Math Derivations

938 Q&As

Structured explanations of assumptions, failure cases, and interviewer follow-ups by module.

Browse Theory Q&A➔

📊DS Cases & Experimentation

586 Guides

Metric drop trees, A/B test sizing, statistical pitfalls, and product analytics case answers.

Browse DS Coach➔

🧬Systems, Cards & Algorithms

477 Cards · 973 Algos

Production AI system designs, roadmap concept cards, coding bridges, and algorithm bank.

Browse Registry➔
🧭Role-Path Curriculum × 3-Layer Flashcards (3,240 Unified Cards Free)

Select Current & Target Role — Drill End-to-End in Order

Automatically skips redundant basics and orders your deck across Bridge, Theory, Coding, System Design & Evals.

🚀Start Path (352 Cards)
📌 Priority 16-Dim Gaps (~18d):D12 L2 SysDesignD01 Serving/ToolingD06 S3 Model/AgentD16 HubB Roles/DSA
🎛️ Open Free Mix Workbench ➔
1LAYER 1 · PROMPT & INTUITION
Science Coding · Kernel CodingHard

FlashAttention Online Softmax Kernel

Implement the block-wise Online Softmax algorithm from FlashAttention-2. Explain how it dynamically rescales intermediate max values and sum denominators without requiring a global reduction over the entire sequence.
💡
Anchor: Block-wise rescale: scale the running sum and output accumulator by exp(m_old - m_new) whenever a new local max is found!
📐
Layer 2: Formula & Math

Active Recall: Mentally formulate the key math & complexity before sliding.

🖱️ Hover to reveal
💻
Layer 3: Benchmark Code

Includes Python reference code, runnable assertions & interview pitfalls.

🖱️ Hover to reveal
Shortcuts:Space Reveal / NextJ / → NextK / ← PrevM MasteredR Random/ Search
Session 0m · 1 Viewed
🗺️Open Full Workbench: All 400 Ordered Cards & Custom Filter Mix ➔

© 2026 TalentMe. All rights reserved.

Designed for Machine Learning Engineers. Powered by Next.js & FastAPI.