English · Español
Extension Module X5 — Interview Prep: Whiteboarding & Systems Design for AI Roles¶
Requires: 40 — Hardening, Postmortem, "What's Next" Teaches:
whiteboarding·systems-design·paper-reading·storytellingJump to any chapter from the phase reference index.
Chapter map¶
🇪🇸 Módulo de preparación para entrevistas en laboratorios de IA: whiteboard ML, systems design para LLMs, lectura rápida de papers, drills de código, behavioral STAR y prep específico por empresa. Convierte el viaje del
lynx-cortexen respuestas listas para usar.
Status¶
- Track: Extension (parallel to core 40-phase curriculum)
- Authorization: Addendum A15 (extension tracks authorized)
- Prerequisites: No hard prereqs — drills cross-link to specific phases.
- Scope guard: Every exercise references the §A13 grammar-tutor artifacts; no novel scope.
- Hardware bar: Pen + whiteboard + 60 minutes per drill. Asciinema-recordable.
Why this module exists¶
The core 40-phase curriculum makes you able to do the work. This module makes you able to talk about doing the work under interview conditions: 60 minutes, hostile follow-ups, a hiring bar tuned for "depth signal".
Most public interview guides are generic. This one anchors every exercise to a PHASE_NN_REPORT.md you already have (or will have) — so your answers carry the weight of receipts, not vibes.
The three signals AI labs actually grade on:
- Depth. Can you go three levels deep without flinching? (theory/01 follow-up trees.)
- Implementation. Can you write attention from scratch in 20 minutes? (theory/04 drills.)
- Systems intuition. Can you do capacity math on a napkin? (theory/02 prompts.)
Module map¶
| File | Topic |
|---|---|
theory/00-interview-landscape-2026.md |
The 2026 AI-lab interview loop: phone screen, ML-systems, coding, paper read, design, behavioral |
theory/01-whiteboard-ml-questions.md |
25 classic ML whiteboard questions with 3-paragraph answers + 3-level follow-up trees |
theory/02-systems-design-for-llms.md |
5 LLM systems-design prompts with capacity math, failure modes, cost discipline |
theory/03-paper-read-drill.md |
How to read a paper in 20 minutes: Attention, CLIP, DPO, Chinchilla |
theory/04-coding-drills.md |
12 coding exercises (attention, BPE, LoRA, DPO loss, RoPE, ...) with 30-min budgets |
theory/05-behavioral-and-storytelling.md |
STAR template + 10 anecdotes pre-filled from the lynx-cortex journey |
theory/06-company-specific-prep.md |
Anthropic, OpenAI, DeepMind, Brain, xAI, Cohere/Mistral signals |
lab/00-mock-interview-checklist.md |
60-minute self-interview drill, 5 × 12-minute questions, asciinema-recordable |
lab/01-paper-pitch-cards.md |
17 elevator-pitch cards: title |
Cross-links to core curriculum¶
Every drill points to the phase that prepared the learner for it. A non-exhaustive map:
- Attention whiteboard / coding drill → Phase 15
- KV-cache systems prompt → Phase 22
- LoRA coding drill → Phase 28
- DPO loss coding drill → X3 Module
- Continuous batcher drill → Phase 33
- Behavioral "complex system" → the 41-phase journey itself
- Behavioral "hard debug" → Phase 19 — Training Dynamics
- Behavioral "tradeoff" → §A13 scope pivot (
LYNX_CORTEX_ADDENDUM.md)
Definition of Done¶
- All 25 whiteboard questions rehearsed aloud at least once; depth-tree traversal verified.
- All 12 coding drills solved within their 30-minute budgets at least once.
- All 17 paper-pitch cards memorized to the 3-numbers-to-remember level.
- All 5 systems-design prompts run end-to-end with capacity math on paper.
- One full asciinema recording of the lab/00 mock interview.
- All 10 STAR anecdotes pre-written and timed to ≤90 seconds each.
Bilingual policy¶
Each file opens with a 1-2 line > 🇪🇸 summary. Body is English-only per project policy.
Further reading¶
Optional — enrichment, not required to pass the phase.
- 📕 Machine Learning Interviews Book — Chip Huyen · 2021. the structure of ML hiring loops.