JoshithReddyAleti/AI_Engineer_Interview_Prep

What it claims

The repo has 811+ interview questions for AI Engineer roles, split into 11 weekly folders that match the 11 episodes of the AI Engineering Roadmap 2026 LinkedIn newsletter. The page says the series is complete: 11 episodes, 11 repos and the full question set. A separate Extra Prep Questions file adds 70+ deep-dive questions on topics that span several weeks:

• advanced RAG variants, embeddings, vector DBs
• agents, fine-tuning, prompt engineering, evaluation
• enterprise security, inference optimization, observability, multi-tenancy

How it works

Each week folder holds 4 prep files. The page does not list what the four files cover. Week 11 adds a fifth file, 05_Terminology_Differentiation.md, which separates easily confused terms:

• chunking vs chunk relevance vs context utilization vs Hit@K
• types of drift
• latency terms
• evaluation methods

Questions include "What the interviewer is really testing" notes and follow-up questions. The page also mentions three features the summary does not:

• Difficulty ratings, meant to calibrate questions for junior vs senior roles.
• System design questions, sized for 45-minute whiteboard rounds.
• Coding questions, usable as standalone exercises.

Each week is paired with its own separate GitHub repo, described as "production-grade", so you can build alongside the questions. Those companion repos are not linked or named in the fetched text.

How to use it

The README suggests different starting points depending on who you are:

• Candidates: start with your weakest week and answer each question before reading the answer. Practise the follow-ups. The README calls the Week 11 terminology file a must-read, because "what's the difference between X and Y?" probes are common.
• Hiring managers: use the files as a screening question bank. Pick questions by difficulty rating, and use the system design questions for 45-minute rounds.
• Self-studiers: pair each week's prep file with that week's companion repo.

Limits

• No licence stated. The fetched page names none. Check the repo before reusing questions in a company screening process or internal material.
• Not validated. Neither the question count nor the "MAANG-level" framing is backed by any evidence on the page, such as sourcing from real interviews or review by interviewers.
• Not a course. The author says so explicitly. The page also does not show answer quality or depth.
• Hard to evaluate from the README alone. The four per-week file types and the companion repos are not identified, so you have to open the repo to judge coverage of any one topic.