Skip to content

Latest commit

 

History

History
84 lines (61 loc) · 6.42 KB

File metadata and controls

84 lines (61 loc) · 6.42 KB

📅 Study Plans

Three plans depending on how much runway you have. All of them assume ~2 hours/day on weekdays and a longer weekend block. Whichever plan you pick, the method is the same:

  1. Read the crash course (README.md) in each topic first - it's the compressed theory.
  2. Self-quiz with questions.md - read the question, answer out loud before opening the collapsible answer. Speaking your answers is the single highest-leverage habit in interview prep.
  3. Type out the coding challenges yourself - don't read the solutions first. The interview is a blank editor, not a multiple-choice test.
  4. Practice system design on a whiteboard or doc, talking through the framework before checking the case study.
  5. The night before any interview: CHEATSHEET.md.

🔥 1-week cram (interview on the calendar)

Triage plan. Skip depth, maximise coverage of what's most likely to be asked.

Day Focus Material
1 LLM fundamentals 02-llm-fundamentals crash course + Basic/Intermediate questions
2 RAG + prompting 04-rag-and-retrieval + 03-prompt-engineering-and-context crash courses, skim questions
3 Agents + evals 06-agents-and-tool-use + 07-evaluation-and-observability crash courses + Basic questions
4 Coding reps 12-coding-challenges: 01 attention, 03 sampling, 08 mini-RAG - implement before peeking
5 System design 11-ai-system-design framework + the case study closest to the company's product
6 Production + safety 08-inference-and-production Basic/Intermediate + 09-safety-security-and-responsible-ai crash course
7 Simulate + rest 13-interview-process-and-behavioral - prep 5 STAR stories; evening: CHEATSHEET.md only

Skip if you must: 10-multimodal (unless the role touches vision/audio), Advanced questions everywhere.


🎯 4-week standard plan (most people)

One theme per week; coding challenges spread throughout so implementation skills compound.

Week 1 - Foundations & the model

Week 2 - Context: prompting, RAG, fine-tuning

Week 3 - Agents, evals, production

Week 4 - Design, safety, polish


🏗 8-week deep plan (career transition into AI engineering)

Weeks 1-4: same as the 4-week plan, at half pace - and build while you learn:

  • After Week 2's material → build a small RAG app over your own notes/docs with an eval harness (even 30 golden questions). This single project teaches more than any tutorial.
  • After Week 3's material → add an agent with 2-3 tools to it, plus tracing.

Weeks 5-8:

Week Focus
5 Depth: re-do every Advanced section across topics 02, 04, 05, 06, 08. Read 5-6 foundational papers from resources (Attention, InstructGPT, LoRA, DPO, ReAct at minimum).
6 Projects: polish one portfolio project to "shows evals + error analysis + tradeoff writeup" standard (see project ideas in 13-interview-process-and-behavioral).
7 System design: all 8 case studies in 11-ai-system-design as timed mocks. All 13 coding challenges done cold.
8 Interview simulation: mock loops with a friend, or alone with the self-mock protocol; behavioural stories rehearsed; company-specific research; CHEATSHEET.md passes.

Retention tips

  • Spaced repetition beats rereading. Second pass on a topic 3-4 days after the first, third pass a week later. The questions.md files are already flashcard-shaped - question first, answer hidden.
  • Track your misses. Keep a running list of questions you fumbled; re-quiz only those on later passes.
  • Explain to a human (or a rubber duck). If you can't explain the KV cache to a non-ML friend, you don't own it yet.
  • Do the numbers by hand once. GPU memory maths, KV cache size, cost-per-request token maths - each done once on paper sticks forever.