Skip to content
Deep-Agnt
PremiumSign in

< Curriculumthe path

The path.

Every section's core problems and projects, in the order to do them. Continue opens the one you are up to.

4 sections · 45 problems · 1 project

0 of 46 done

Start the curriculum Next: 01 Make Your First Model Call

Section 01 · Foundations

Model APIs

0 of 15

The model as a function you call, and everything that goes wrong when you do.

Part 01 Call the model

  1. 01Make Your First Model Callimplement · easy
  2. 02Carry the Conversation Yourselfdebug · easy
  3. 03Format a Conversation With a Chat Templateimplement · medium
  4. 04Round-Trip Text Through a Tokenizerimplement · easy
  5. 05Price a Callimplement · easy

Part 02 Get an answer you can use

  1. 06Get the JSON Out of a Model's Replyimplement · easy
  2. 07Validate LLM Structured Outputimplement · easy
  3. 08Ask Again When the JSON Is Wrongimplement · medium
  4. 09Handle the Reply That Got Cut Offimplement · medium
  5. 10Sample the Next Tokenimplement · medium

Part 03 Streams and failures

  1. 11Assemble a Reply From a Streamimplement · medium
  2. 12Parse JSON From a Streamimplement · medium
  3. 13Retry a Failed Model Callimplement · easy
  4. 14Batch Calls Without Tripping the Rate Limitimplement · medium

Part 04 Build

  1. 15model-clientproject

Section 02 · Foundations

Context engineering

0 of 9

What the model sees is all it knows. Decide what goes in, in what order, within a budget.

Part 01 What goes in

  1. 16Fix the System Prompt That Leaksdebug · medium
  2. 17Pick the Examples That Fit the Questionimplement · medium
  3. 18Put the Answer Where the Model Will Read Itimplement · medium

Part 02 Within a budget

  1. 19Trim the Conversation, Keep the System Promptimplement · medium
  2. 20Stay Under the Token Budgetoptimize · medium
  3. 21Compact a Long Conversation Without Losing the Decisionimplement · medium
  4. 22Lay Out the Prompt So the Cache Hitsoptimize · medium

Part 03 Room to think

  1. 23Ask Five Times, Keep the Majorityimplement · medium
  2. 24Answer Over a Context That Does Not Fitimplement · hard

Section 03 · Foundations

Retrieval

0 of 12

Find the right text before you ask. Most bad answers are bad retrieval.

Part 01 Embeddings

  1. 25Calculate Embedding Similarityimplement · easy
  2. 26Generate Embeddings with Hugging Faceimplement · medium
  3. 27Build Top-K Semantic Searchimplement · easy
  4. 28Search by Meaning, Not by Wordsimplement · medium

Part 02 Keywords and fusion

  1. 29Build BM25 From Scratchimplement · medium
  2. 30Fuse BM25 and Dense Search With RRFimplement · medium
  3. 31Rewrite the Query Before You Searchimplement · medium
  4. 32Filter by Metadata Before You Rankimplement · medium

Part 03 Chunk, rerank, measure

  1. 33Build a Document Chunkerimplement · easy
  2. 34Rerank the Retrieved Chunksimplement · medium
  3. 35Rerank With a Model That Understands the Questionimplement · medium
  4. 36Measure Retrieval with Recall@k and MRRimplement · medium

Section 04 · Foundations

RAG and grounding

0 of 10

Answers that cite what they used, and say so when nothing supports them.

Part 01 Retrieve, then generate

  1. 37Build Your First RAG Pipelineharness · medium
  2. 38Fix a Broken RAG Systemdebug · hard
  3. 39Answer With Citations, or Say You Can'timplement · medium
  4. 40Check Every Sentence Has a Sourceimplement · medium

Part 02 Say so when nothing supports it

  1. 41Make the Model Admit It Doesn't Knowimplement · hard
  2. 42Know When Not to Retrieveimplement · medium
  3. 43Decide When Two Documents Disagreeimplement · medium

Part 03 Beyond one search

  1. 44Answer the Question That Needs Two Documentsimplement · hard
  2. 45Hand Retrieval to the Model as a Toolimplement · medium
  3. 46Answer From a Repo That Does Not Fitimplement · hard
Back to the curriculum