All learning paths

Course outline

Intermediate

Build real AI apps with transformers, RAG, structured outputs, agents, MCP, workflows, and evals

7-9 weeks

Self-paced

44

Lessons

9

Modules

1

Foundation of Transformers

6 lessons

1Which problem did each generation of NLP solve?

2Read the transformer as a set of design decisions

3Calculate attention, then change what it can see

4Which information can an encoder and decoder use?

5How does a transformer know which token came first?

6What changes when every task becomes text in and text out?

2

Modern LLM Architectures

5 lessons

1What changed between text completion and an assistant?

2What can a model learn by filling gaps in a sentence?

3When does extra reasoning improve the result?

4Follow a causal decoder from prompt to next token

5Choose an encoder output that matches the task

3

LangChain & Frameworks

5 lessons

1What should a framework simplify, and what must stay visible?

2Who chooses the next step: your code or the model?

3What should a conversation remember, and for whom?

4Build a small answerer with a visible evidence boundary

5Build an agent that can say it has not finished

4

RAG (Retrieval-Augmented Generation)

5 lessons

1Can you make an answer depend on the right evidence?

2What does a vector database store besides vectors?

3What does a similarity score really compare?

4Build the retrieval path before adding a larger stack

5Build a document helper whose answers can be audited

5

Vector Databases & Embeddings

5 lessons

1Why do some embeddings work for search and others do not?

2Compare vector stores with your workload, not a ranking table

3Where should you split a document without breaking its meaning?

4How do you combine exact matches with semantic matches?

5Build search you can inspect before adding answers

6

Agentic AI Frameworks

5 lessons

1Agents: who chooses the next step?

2Google ADK: build a small tool-using agent

3OpenAI Agents SDK: inspect the runner loop

4CrewAI: separate tasks without losing evidence

5Project: compare agent frameworks on one task

7

MCP Connectors & Workflows

5 lessons

1MCP connections: follow one request across the boundary

2Build a read-only MCP server

3Workflows: make every step and failure visible

4LangGraph: follow state through a graph

5Project: a workflow that survives an interrupted session

8

Modern AI Development

5 lessons

1MCP messages: discovery, calls, and results

2LlamaIndex: inspect what enters the index

3AI SDK: streaming without pretending the answer is finished

4AI development tools: investigate a change before accepting it

5Project: a source-backed assistant that remembers its user

9

Application Reliability

3 lessons

1Structured output: valid shape, uncertain facts

2Context engineering: choose what the model needs now

3Evaluation: find out what actually improved