Craft with AI · interactive course · 42 lessons
42 lessons across 7 modules — Python to LLMOps — taught the way modern engineers actually work: alongside an AI assistant, with code you run in your browser. One command opens the whole course.
# the entire course, one command
$ ./start_linux.sh
✓ python 3.10+ detected
✓ virtualenv ready .venv
✓ dependencies installed
✓ integrity check license · 42 lessons
→ serving course http://127.0.0.1:8765
# your browser opens to a live, editable,
# runnable syllabus — progress auto-saved.Curriculum
The modules assume zero prior knowledge — lesson one is “Hello, World!” — and end on zero-downtime cloud deployment. Each lesson is a runnable Python file with a difficulty, a stated goal, and a “Why this matters for AI” section.
Variables, data types, operators, strings, control flow, loops, lists, dictionaries, tuples, sets, functions, modules & packages — then real APIs, pandas data analysis, and a practical mini-project to close it out.
A weather-data tool from a live API, and a pandas/matplotlib study that plots real dependencies.
The toolchain layer most bootcamps skip: the Linux shell, file paths and permissions, Git local & remote with merge-conflict rescue, virtual environments, and packaging a codebase for LLM context.
A codebase packager that shrinks a repository into an AI-ready context prompt.
The HTTP protocol, Pydantic validation, async/await concurrency, and FastAPI routers with auto-generated docs — the same stack production backends use.
A FastAPI backend with validation, CORS, and live /docs.
“Works on my machine” solved for good: containers vs VMs, writing efficient Dockerfiles, the Docker CLI, and multi-service orchestration with Compose.
A Dockerfile auditor, and a Compose stack wiring FastAPI to a vector database.
Resilient LLM API clients with retry and backoff, structured JSON outputs with a self-correction loop, chunking & embeddings, and vector search in ChromaDB.
A working RAG retriever: documents in, top-K semantically-relevant hits out.
Agents without the framework bloat: LLM tool calling, the ReAct loop (Thought → Action → Observation), and stateful memory management.
A ReAct loop engine from scratch, plus a stateful agent scratchpad.
The last mile: tracing and telemetry for LLM pipelines, CI/CD quality gates with GitHub Actions, and cloud deployment with zero-downtime releases.
A JSON pipeline tracer, a CI audit parser, and a rolling-update simulator.
Method
The course is engineered around how building software actually works today — and around how people actually learn. Three decisions carry it.
Every lesson is a real, runnable Python script — there is no separate course database. Each code block gets a “Try it here” panel with a 4-second sandboxed runner, spotlight search, day/night mode, and progress that auto-saves.
A dedicated workspace module teaches codebase packaging, token bloat, and merge-conflict rescue — the skills that make Cursor, Copilot, and Gemini genuinely useful instead of occasionally wrong.
Every lesson opens with why. Real-world analogies, “In simple words” recaps, and “Before you run it” notes turn syntax into mental models you can reuse when orchestrating whole systems.
Portfolio
No toy exercises. Across 42 lessons you assemble a small portfolio of working tools — it starts as plumbing, ends as a deployable AI system.
On the record
Everything below is measured from the actual lesson files — no inflated “60-hour course” claims.
16h 25m.985 min
# === header in the source.464
248
5 levels
127.0.0.1.std-lib only
gated
Start here
No account, no hosted platform, no install of heavyweight tools. Python 3.8+ is the only prerequisite on any machine.
# one script sets up venv, deps, and opens the course
$ git clone https://github.com/prawinin/craft-with-ai.git
$ cd craft-with-ai
$ chmod +x start_linux.sh
$ ./start_linux.sh
# Windows: double-click start_windows.bat
curl → http://127.0.0.1:8765 # your browser, already thereDev-workspace skills start before the first lesson — you use Git and the shell the whole way through.
The start script creates a virtualenv, installs only the handful of real deps, and verifies course integrity.
The course opens on localhost with search, day/night mode, and progress tracking from the first lesson forward.
Craft with AI · free to learn
42 lessons will not make a course for you. Reading them with the code running in front of you might just make an AI engineer.