AI Engineer Foundations

The plan, week by week

Times are each lesson’s own reading time, added up. A mock interview is set to 45 minutes. A lock means the lesson is in the course.

1. Week 1

   How a model reads and writes

   300 min

   Know what goes into a model and what comes out.

   1. Why ML Models Fail in Production: The Production GapAI Engineering lesson55 minfree
   2. What a Token Is: How a Language Model Reads TextAI Engineering lesson51 minin the course
   3. What an Embedding Is: Meaning as a List of NumbersAI Engineering lesson48 minin the course
   4. What a Language Model Actually Outputs: Odds for Every Next TokenAI Engineering lesson50 minin the course
   5. Temperature and Sampling: How One Token Gets PickedAI Engineering lesson50 minin the course
   6. The Context Window: What Happens When a Prompt Does Not FitAI Engineering lesson46 minin the course
2. Week 2

   Prompting

   291 min

   Ask for exactly what you want, in a shape your code can read.

   1. The Parts of a Prompt: Adding One Piece at a TimeAI Engineering lesson45 minin the course
   2. Vague, Soft and Specific: Asking for Exactly What You WantAI Engineering lesson47 minin the course
   3. Asking for a Format: Words, JSON Mode and a SchemaAI Engineering lesson45 minin the course
   4. Examples in a Prompt: How Many, Which Ones, in What OrderAI Engineering lesson47 minin the course
   5. Quoting Untrusted Input: Tags, a Reminder, and a Fake Closing TagAI Engineering lesson52 minin the course
   6. Testing a Prompt on New Cases: Why the Score You Tuned On May Be OptimisticAI Engineering lesson55 minin the course
3. Week 3

   Retrieval and RAG

   332 min

   Find the right pieces of your documents for each question.

   1. Searching by Meaning: Real Semantic Search, in Six LanguagesAI Engineering lesson55 minin the course
   2. Splitting Text Into Pieces: Why Search Works Better on Small ChunksAI Engineering lesson47 minin the course
   3. Chunking: The First Lever on Retrieval QualityAI Engineering lesson60 minin the course
   4. Hybrid Retrieval: When Keyword Search Beats Your EmbeddingsAI Engineering lesson60 minin the course
   5. Reranking: The Highest-Return Change in Your RAG StackAI Engineering lesson55 minin the course
   6. The RAG Scale Cliff: What Breaks Between 100 and 5 Million DocumentsAI Engineering lesson55 minfree
4. Week 4

   Evals

   367 min

   Measure answers instead of guessing.

   1. Why Evals Are the Job: Moving From Vibes to MeasurementAI Engineering lesson47 minfree
   2. Error Analysis: Reading Failures Before Building MetricsAI Engineering lesson55 minin the course
   3. Building a Golden Dataset: The Set That Catches RegressionsAI Engineering lesson55 minin the course
   4. LLM-as-Judge and Its Failure ModesAI Engineering lesson65 minin the course
   5. Validating Your Judge: TPR, TNR, and the Human CeilingAI Engineering lesson60 minin the course
   6. RAG Evaluation: Localizing Failure Between Retrieval and GenerationAI Engineering lesson85 minin the course
5. Milestone

   A RAG baseline and an eval harness

   129 min

   Design a full RAG system, then defend it in a mock interview.

   milestone: A RAG baseline and an eval harness that tells you when it gets worse.

   1. Design RAG Systeminterview guide, free14 minfree
   2. Design a RAG System at ScaleSystem Design lesson70 minin the course
   3. Mock interview: design RAG SystemAI interviewer, mid-level level45 min

Questions about this path

Do I need machine learning experience for this path?

No. You need to be able to write Python and call an API. The first week explains what a token, an embedding and a model's output really are, in plain words.

What is a RAG baseline?

RAG means the system first searches your documents, then gives the best pieces to the model so it answers from them. A baseline is the first simple version you measure, so every later change can be compared with it.

What is an eval harness?

A set of test questions with known good answers, and code that scores the system on them. It tells you, with numbers, whether a change made the answers better or worse.

Other paths