Five courses. Do the Foundations course first; Agent Memory, Context Compression, Feedback Loops, and Creating MCPs are siblings that each assume it and build on top — read those in any order.
01
AI Development — Foundations
Start here. Build from a raw HTTP request up through tokens, sampling, tools, the tool-use loop, sandboxing, embeddings, RAG, and agents — seeing the raw mechanic first, then the convenient abstraction on top.
24 sectionsread in orderStart here
Start the course →
02
Agent Memory: From Chat History to a Knowledge Graph
Do Foundations first. Goes deep on cross-session memory: a taxonomy of memory, conversational ingestion, a Neo4j knowledge graph, hybrid retrieval, lifecycle and decay, measurement, and a measured, opinionated default.
12 unitsread in orderAssumes Foundations
Continue the course →
03
Context Compression: Keeping a Long Agent Inside the Window
Do Foundations first. A sibling of Agent Memory: keeping one long session inside the token budget without losing what matters — measuring the window, dropping and windowing, structured summarization, head/middle/tail preservation, a deterministic pre-pass, and a measured default the instrumentation earns.
13 unitsin progressAssumes Foundations
Read the course →
04
Feedback Loops: Earning Autonomy by Being Observable
Do Foundations first. Observability is the subject here: bubble up the telemetry that makes every harness and model decision visible, then climb the autonomy gradient — runtime gates, self-reflection, human-in-the-loop, meta-monitoring — to feedback loops you can trust. Don’t ship a black box.
13 unitscompleteAssumes Foundations
Read the course →
05
Creating MCPs: Safe, Observable, Bounded Servers for the Enterprise
Do Foundations first. Build MCP servers over a real CMDB (Postgres) and SIEM (OpenSearch): catalogs and discovery, log and incident search, restrictive vs permissive tools, identity and enterprise rules, per-caller limits, performance, the OWASP MCP Top 10, and audit. Labs in marimo.
13 unitsin progressAssumes Foundations
Start the course →
You write every line
Each lesson guides you through building small scripts yourself. A runnable reference solution
for everything lives under examples/.
One hosted endpoint
Every example targets an OpenAI-compatible server (vLLM serving gpt-oss-120b). Point three
environment variables at it and go; embeddings can use a separate endpoint.
Security throughout
Security is a through-line, not a chapter — every lesson ends with a topic-specific note, and isolation gets dedicated sections.