Start where you are — each level maps to a different point on the journey. We confirm the right one together in your free first session.
01$90/hr
LLM Foundations
New to building with LLMs? Start here.
Who it's forFor confident Python users who use ChatGPT daily and now want to call LLMs from their own code and ship a first working app — no ML background required.
Go from ChatGPT user to builder. Learn how large language models actually work, call the OpenAI and Claude APIs from Python, get reliable structured…
What you'll be able to do
Call the OpenAI and Anthropic (Claude) APIs from Python and handle responses, errors, and retries.
Get back valid, schema-conforming JSON every time, using response formats / tool schemas.
Estimate token usage and cost before you send a request, and stay inside the context window.
Build and run a small working chat app (CLI or Streamlit) that holds a multi-turn conversation.
Sounds familiar?
I can prompt ChatGPT fine, but I have no idea how to actually call it from a Python script.
My API calls work once, then break — and I don't understand tokens, context limits, or why I'm getting charged.
I asked for JSON and got a paragraph with JSON buried in it; I can't parse the output reliably.
There are OpenAI and Claude and a dozen tutorials — I don't know which to start with.
LLM & token basicsOpenAI / Claude APIPrompting & structured outputYour first chat app
Who it's forFor someone who can already call an LLM API and wants to ground it in their own data and let it take actions — building retrieval (RAG) and tool-using agents.
Build the systems behind real AI products: retrieval-augmented generation over your own data, vector databases, and tool-using agents with…
What you'll be able to do
Embed a document set, store it in a vector database, and retrieve the right chunks for a question.
Build a working RAG pipeline that answers questions grounded in your own data, with sources.
Implement an agent that calls tools (search, a function, an API) and loops until the task is done.
Structure an agent as a graph in LangGraph and reason about its state, branches, and failure points.
Sounds familiar?
The model makes things up about my domain — I need it to answer from my documents, not its training data.
I keep hearing "embeddings" and "vector database" but I don't understand what to store or how retrieval works.
My RAG returns irrelevant chunks, so the answers are still wrong — chunking and retrieval quality are a mystery.
I want the LLM to call a function or API, but tool-calling and agent loops feel like magic I can't debug.
Who it's forFor someone with a working LLM prototype (RAG or agent) who now needs to make it trustworthy in production — evals, guardrails, cost and latency control, and a clear fine-tune-vs-prompt decision.
Take an LLM prototype to production. Design evaluations you can trust, add guardrails, decide when to fine-tune (LoRA/QLoRA) versus prompt, and…
What you'll be able to do
Design an eval set and run automated tests (including LLM-as-judge) so changes are measured, not guessed.
Add input/output guardrails that catch unsafe or malformed responses before they reach users.
Run a small LoRA/QLoRA fine-tune and judge, with data, whether it beats a strong prompted baseline.
Cut cost and latency with caching, model routing, and streaming, and monitor quality in production.
Sounds familiar?
My demo works, but I have no way to know if a prompt change made it better or quietly worse.
It occasionally says something unsafe and I have nothing catching that before the user sees it.
Everyone says "just fine-tune it" — I can't tell when that's worth it versus better prompting or RAG.
It's too slow and the bill is scary; I don't know where the latency or the tokens are going.
No guarantees, no fixed curriculum — just a specific, repeatable way of working that gets you unstuck on AI Engineering.
01
Built around your goal
There is no fixed syllabus to keep pace with. The hour is built backwards from the one thing you need — a failing assignment, a concept that will not stick, a project to ship.
02
Diagnosed, not re-taught
We find the precise step where it breaks down instead of re-covering what you already know — so the time goes to the gap that actually matters.
03
You drive, I steer
You do the work in real time while I guide — that is how it sticks. You leave able to do it yourself, not just having watched me do it.
04
Honest pace & pricing
You only pay for the levels and pace that fit. We agree the plan together after the free first session — no packages you do not need.
// faq
Frequently asked questions
About AI Engineering tutoring and how sessions work.
Is the first AI Engineering session really free?
Yes. Your first session is complimentary so you can experience the teaching style, talk through your goals, and decide whether to continue — no credit card required upfront.
How much does AI Engineering tutoring cost?
Sessions start at $90/hour, and multi-session packages are available at a discount. You only pay for the levels and pace that fit your goals — we agree on a plan together after the free first session.
How are AI Engineering sessions delivered?
All sessions are 1-on-1 and 100% online over video, with screen sharing and a shared editor or whiteboard. Sessions are typically 60–90 minutes and scheduled around your availability.
Which AI Engineering level should I start at?
It is set by where you are now, not a fixed curriculum. In the free first session we map your background to the right starting level and adjust the pace as you progress.
Who is teaching the sessions?
Every session is taught directly by Ali Jabbary, M.Sc., P.Eng. — not a rotating pool of tutors. You work with the same instructor throughout.
