An open curriculum · built and verified in the open
AI Engineering.
The field guide — every section, course, lecture, and answered question in this repository, organized so a complete beginner can find the way in.
243 study notes
6 curriculum sections
19 course lectures
100 interview Q&As
251 indexed files
This repository is a free, self-study library for becoming an AI engineer, written from first principles so you can start from zero. It has three layers that work together: a technical curriculum of six numbered sections (101 LLM Literacy through 106 AI Security & Safety) that build on each other in learning order; roadmaps — guided day-by-day study plans through that same material, from a 286-session full roadmap to a 100-day accelerated plan; and interview & career material, including a 100-question bank where every question carries a worked expert answer. Inside section 105 sits a complete 19-lecture course on LLM evaluations — every lecture transcribed from Hindi to English, expanded into structured notes, and cross-linked to research-grade reference material. What it does not pretend to be: a paid bootcamp, a substitute for building things, or a source of unverified numbers — the evidence section keeps the discipline at the front door.
The learning chain — the six sections in the order they build on each other
101 understand→102 build→103 automate→104 operate→105 prove→106 defend
Study notes243
Curriculum sections6 units 101–106
Course lectures19 + 19 transcripts
Glossary terms120 linked to lectures
Quiz questions60 verified answers
Interview Q&As100 all answered
Roadmap sessions286 full + 100-day
Portfolio projects12 with done-when criteria
Files indexed251
Verified repo links416 0 broken
Every number above is generated — this page is rebuilt by tools/build_site.py from the repository's own file tree, and every GitHub link on it is checked against that tree at build time. The content numbers (120 glossary terms, 60 quiz questions, 286 sessions) are the repository's own verified figures from its section READMEs, last audited October 27, 2026.
§01 · Beginner onboarding
New here? Start with this section
The repository has three layers. You don't need all of them at once — pick the layer that matches your situation, and open the others when you need them.
Layer 1 · The knowledge
Technical curriculum, 101–106. Six numbered sections — LLM Literacy → Application Development → Agentic AI → LLMOps → Evaluation → Security & Safety — written from first principles: no machine-learning background required. The curriculum ↓
Layer 2 · The schedule
Roadmaps. Guided, day-by-day study plans through the same material: a 286-session full roadmap, a 100-day accelerated plan, 100 interview questions, and 12 portfolio projects. The roadmaps ↓
Layer 3 · The job
Interview & career material. The AI-engineer role explained, learning paths from five adjacent careers, a portfolio guide, and interview preparation where every question carries a worked answer. Interview prep ↓
Prerequisites — you need less than you think
You can already program (Python is the most useful language here). That's it.
No machine-learning background is required — section 101 teaches every concept
from scratch, and every later section links back to what it builds on.
Which part should you open first?
| Your situation | Start here | What it is |
|---|---|---|
| "I'm completely new to LLMs and AI engineering" | 101 — LLM Literacy | The on-ramp: how LLMs work, prompting, context engineering — from zero |
| "I want a day-by-day plan to become job-ready" | Accelerated Roadmap (100 days) | 6 modules · 100 days · the core ~80% employers expect |
| "I want full depth, at my own pace" | Full Roadmap | 16 levels · 286 sessions · software fundamentals to interview prep |
| "I need to learn one specific topic" | Curriculum sections 101–106 | Reference knowledge base — open just the section you need |
| "I'm switching from another engineering role" | Learning Paths | Transition guides from data / backend / frontend / ML engineering |
| "I have an interview coming up" | Interview Preparation | The process, 100 answered questions, system design, negotiation |
Your first 60 minutes — a concrete first session
Read the repository's own introduction~10 min
Open the main README and skim how the three layers fit together. Don't follow every link yet — just get the map.
Read your very first technical note~15 min
"What Is a Large Language Model?" — the conceptual layer, explained from zero. This is note 1 of the whole curriculum; if you understand it, you're ready for everything that follows.
See where you're headed~10 min
Skim Module 1 of the Accelerated Roadmap (LLM fundamentals, days 1–15). You won't know these terms yet — that's the point: it's a preview of the trail, not today's climb.
Locate your personal entry point~25 min
If you already work in tech, read the transition guide for your current role in Learning Paths — data engineer, data scientist, ML engineer, backend, frontend, each with a duration estimate. Coming from outside tech? Follow the Full Roadmap, which starts at software fundamentals.
Roadmaps versus curriculum — how they fit together
Roadmaps = the guided tour
They tell you what to study and in what order — your schedule. The full roadmap covers 100% of the field in 16 levels; the accelerated one compresses it to the core ~80% in 100 days. Follow a roadmap as your plan.
Curriculum = the reference library
Deep notes on each topic, organized 101–106. Whenever a roadmap mentions a topic you want in full, open the matching curriculum section. Use them as your dictionary while you build.
§02 · Layer 1, the knowledge
The technical curriculum — sections 101–106
Six numbered sections in learning order. Each subsection's notes are numbered — open the folder and read them 1, 2, 3… Every link below goes to GitHub, where the notes render as formatted Markdown.
101
LLM Literacy
First-principles guide to transformer mechanics, inference sampling, prompt & context engineering. The on-ramp for absolute beginners — no ML background needed.
19 notes · 3 units
key read: What Is an LLM?
key read: What Is an LLM?
102
AI Application Development
Orchestration, vector databases, the complete RAG curriculum, memory systems, fine-tuning, and AI automation — the everyday craft of the AI engineer.
41 notes · 6 units
key read: FT vs RAG vs Prompting
key read: FT vs RAG vs Prompting
103
Agentic AI
Autonomous planning loops, tool protocols (MCP), agentic frameworks, and multi-agent orchestration — and when not to add an agent.
18 notes · 4 units
key read: Intro to MCP
key read: Intro to MCP
104
LLMOps
Deployment architectures, observability, safety guardrails, governance, and enterprise scaling — everything between "the demo works" and "the system runs reliably".
35 notes · 6 units
key read: CI/CD for AI Systems
key read: CI/CD for AI Systems
105
AI Evaluation
Evidence: proving your system works. Evaluation fundamentals, LLM-as-a-Judge, agent benchmarks and the integrity crisis, RAG evaluation, eval engineering in CI/CD, dataset engineering, and online experimentation — plus the complete 19-lecture course below.
52 files · 7 notes + course
key read: Evaluation fundamentals
key read: Evaluation fundamentals
106
AI Security & Safety
Threat modeling (OWASP LLM + Agentic Top 10), prompt injection defense, red teaming, agent & MCP security, AI alignment, and responsible AI governance — plus catalogs of real security datasets and real incident databases.
9 notes · 2 catalogs
key read: Threat modeling
key read: Threat modeling
§03 · Inside section 105, the curriculum layer
The 19-lecture LLM Evaluations course
A video-by-video, step-by-step English study companion for the "LLM Evaluations: The Step-by-Step Playlist for 2026" course by CampusX (taught in Hindi by Nitish Singh). Every lecture transcribed, translated to English, and expanded into structured expert notes — with nothing skipped.
What the course teaches — evaluation, from zero to production
The complete testing setup, not just metrics: model vs application evals · the end-to-end
workflow · LLM-as-a-Judge, reference-based vs reference-free · benchmark literacy
(MMLU, GSM8K, SWE-bench…), saturation, contamination, leaderboards · RAG evaluation in
practice (retriever testing, the RAG Triad, G-Eval) · security testing (toxicity,
leakage, scope drift) · operational evals, regression testing, and online monitoring
with LangSmith.
The lectures — notes and transcripts
| # | Lecture | Duration | Study materials |
|---|---|---|---|
| 1 | Master LLM Evaluations — playlist overview | 23:24 | Notes · Transcript |
| 2 | Introduction — Model Evals vs Application Evals | 24:32 | Notes · Transcript |
| 3 | How to Evaluate LLM Applications — the Complete Workflow | 17:00 | Notes · Transcript |
| 4 | Why Your AI Application Needs Multiple Eval Pipelines | 28:06 | Notes · Transcript |
| 5 | LLM Eval Methods — LLM-as-a-Judge, Reference-Based vs Reference-Free | 53:10 | Notes · Transcript |
| 6 | Offline Evals vs Online Evals | 1:21:09 | Notes · Transcript |
| 7 | LLM Model Evals & Capabilities | 37:52 | Notes · Transcript |
| 8 | What is LLM Benchmarking — Saturation vs Contamination | 51:20 | Notes · Transcript |
| 9 | What are LLM Benchmarks — Evolution of AI Knowledge Benchmarks | 1:50:47 | Notes · Transcript |
| 10 | How to Use LLM Leaderboards | 30:07 | Notes · Transcript |
| 11 | Selecting the Right LLM — Running Custom Model Evals | 1:56:53 | Notes · Transcript |
| 12 | How to Answer "How Do You Evaluate Your RAG App?" in Interviews | 46:20 | Notes · Transcript |
| 13 | How to Test RAG Retrievers (Hands-On) | 1:47:02 | Notes · Transcript |
| 14 | Evaluating RAG — Generator & Full Pipeline with the RAG Triad | 1:21:29 | Notes · Transcript |
| 15 | Mastering G-Eval — the Deterministic LLM-as-a-Judge Framework | 1:26:52 | Notes · Transcript |
| 16 | Securing Your RAG Application — Toxicity, Leakage & Scope Drift | 1:35:19 | Notes · Transcript |
| 17 | RAG Operational Evals — Building Faster & Cheaper RAG Systems | 1:19:32 | Notes · Transcript |
| 18 | RAG Regression Testing — Preventing Silent AI Failures | 54:16 | Notes · Transcript |
| 19 | Online Evaluation & Monitoring with LangSmith | — | Notes · Transcript |
The study aids
Glossary — 120 terms
All 120 course terms, defined and linked to the lecture that teaches each. Use it when a term is unfamiliar, or before an interview sweep. Open →
Self-assessment quiz — 60 Q&A
60 questions with verified answers, 15 per unit. Finished a unit? Check retention before moving on — miss more than 2 and re-read. Open →
Cheat sheet
Every formula, number, and decision rule from all 19 lectures on one page. For revising after the course, or right before an interview. Open →
Progress tracker
Per-lecture checklist with quiz checkpoints and time budgets — a structured plan through the 34-hour full path. Doubles as a project plan. Open →
How to study this course
Watch/read in order the first time — the course is deliberately cumulative (the
model-vs-application distinction from lecture 2 powers every later lecture). Test
yourself after each unit with the quiz, look up unfamiliar terms in the glossary, and
follow each note's cross-links into the section-105 reference notes for research-grade
depth. Lecture notes include timestamps like (12:18) and clearly-separated
expert commentary blocks that add context the lecture didn't cover.
§04 · Layer 2, the schedule
Two roadmaps + the interview track
Structure adapted from the CampusX AI Engineering Roadmap 2026, with every learning resource replaced by free worldwide alternatives. Pick one roadmap as your schedule — not both.
Full Roadmap — 100% coverage, own pace
16 levels · 7 phases · 286 sessions · 14 projects — from software fundamentals to interview prep. Start here if you're new to programming or want maximum depth.
- L0 · Software Engineering Fundamentals
- L1 · LLM 101
- L2 · Working with LLM APIs
- L3 · Prompt Engineering
- L4 · RAG
- L5 · AI Agents
- L6 · Context Engineering
- L7 · LLM Evaluations
- L8 · AI Security
- L9 · AI System Design
- L10 · LLMOps
- L11 · Managed AI Platforms
- L12 · Multimodal 101
- L13 · Fine-Tuning
- L14 · Projects
- L15 · Interview Prep
Accelerated Roadmap — job-ready in 100 days
6 modules · 100 days — the core ~80% you need to land your first AI engineering job. Start here if you already program and want the shortest employable path.
- Module 1 · LLM Fundamentals (days 1–15)
- Module 2 · Building with LLM APIs
- Module 3 · Prompt Engineering
- Module 4 · RAG
- Module 5 · Agents
- Module 6 · Shipping to Production
Open the Accelerated Roadmap →
Also in the roadmap folder: the 100 interview questions with strong-answer outlines, the matching full model answers, and 12 portfolio project specs. Not sure which fits? The roadmap hub compares them — including how important each topic is for interviews versus the real-world job.
§05 · Layer 3, the job
Interview preparation
The complete track: how the process works, 100 questions with worked expert answers (every question in the bank carries an answer), system design, take-homes, negotiation, and a resume playbook.
The process
The question bank — all answered
§06 · Layer 3, the job
The role, your path in, and your portfolio
What the AI engineer role actually is, a personalized route from five adjacent careers, and the portfolio that proves you can do the work.
The AI engineer role
Learning paths — pick your origin
- General learning path — what & in what order
- From ML Engineer — easiest transition
- From Data Engineer — 3–4 months
- From Data Scientist — evaluation is your superpower
- From Backend Engineer — 2–3 months
- From Frontend Engineer — backend first, then AI
- Advanced path — foundations → advanced engineering
Portfolio
The highest-leverage 2026 portfolio project: an AI application with its own evaluation suite — golden datasets, RAG Triad metrics, safety evals, regression gates. A RAG chatbot with a visible eval suite beats a fancier chatbot without one.
§07 · Evidence & extras
Research, evidence & resources
This repository separates measurements from claims. The research folder holds the reproduced job-market analysis and the audit trail of everything checked and corrected; the awesome list holds curated external resources.
Research & evidence policy
- Evidence & real datasets — reproduced job-posting analysis
- Evidence policy — measurements vs claims vs targets
- Repository audit — checked, corrected, unresolved
The reproduced source contains 8,051 stored job observations across nine scrape folders (1,086 in the September 23, 2026 folder) — counts from a selected job-board dataset, not worldwide demand estimates. The analysis fingerprint makes counting reproducible.
Awesome AI Engineering
Curated external resources compiled while researching this guide — every link fetched and verified in the October 9, 2026 review: practitioner interview stories, AI system design guides, company engineering blogs, books & courses, case-study collections.
Webinars (supplemental)
- A Day of an AI Engineer — the practical reality
- Defining the AI Engineer Role
- The Interview Process — live hiring trends
- Take-Home Assignments — real examples analyzed
Recording, speaker and survey claims in these notes have not been independently verified — see their evidence status.
The sister repository
AdilShamim8/AI-Safety — the deep
curriculum behind section 106's dataset and incident catalogs: 24 chapters,
10 runnable projects, 47 verified dataset cards with a machine-readable
catalog, 261 registered sources, and worked incident reviews — the evidence
ladder this repository's security claims are checked against. Cross-verified
October 27, 2026. Its own monograph-style site:
adilshamim8.github.io/AI-Safety.
§08 · The complete index
The library — every file in the repository
Every note, transcript, dataset card, script and data file — 251 items, machine-generated from the repository tree, each linking straight to its rendered page on GitHub. Filter by section or search by name.
§09 · Colophon
How this page is made
Design
A single-file editorial site in the same paper-and-ink monograph style as the sister site adilshamim8.github.io/AI-Safety: Georgia on paper, one oxblood accent, hairline rules, no frameworks, no external fonts, no network. Works offline; print styles included.
Generation
Rebuilt by tools/build_site.py from the repository's own file tree. The file index, the ledger counts, and every GitHub link are generated — never hand-typed — and each build verifies that every embedded repository link resolves to a real file.
Verification
- Every commit authored by AdilShamim8 (API-verified)
- References verified October 9 & October 27, 2026
- Dataset figures checked against live primary sources
- MIT License — free forever
generated 2026-10-27 from the AI-Engineering repository tree · 251 files indexed ·
416 repository links verified, 0 broken · single file, no dependencies ·
regenerate: python tools/build_site.py