Academy

First cohort · April 2027

Join the shortlist

FIRST COHORT · APRIL 2027 · BERLIN OR LIVE ONLINE

Build, evaluate and run applications on language models.

A full-time, live program for people who already write code. You learn to build LLM applications the way an engineer builds any system: retrieval, tools, evaluation, deployment, monitoring. Fifteen seats, taught by Alexey Grigorev.

15 SEATS · 1 INSTRUCTOR · NONE OFFERED YET

Taught live by Alexey Grigorev, founder of DataTalks.Club and creator of the free Zoomcamps (85,000+ GitHub stars across the course repositories).

START
April 2027
FORMAT
Full-time, live, 12 to 16 weeks
WHERE
Berlin (Prenzlauer Berg) or live online
LANGUAGE
English
SEATS
15
TUITION
Bildungsgutschein planned (approval pending) or self-pay

01 / WHAT YOU WILL BE ABLE TO DO

By the end you can build an AI application, and show where it works and where it fails.

  1. 01Build an assistant in Python on an LLM API.
  2. 02Use retrieval-augmented generation (RAG) and document processing over company data.
  3. 03Give the assistant tools, function calling and MCP connections to look things up or call an API.
  4. 04Define what it can do on its own and when it needs human approval.
  5. 05Evaluate answers and tool use with realistic tasks; compare prompts, models and retrieval methods; catch regressions.
  6. 06Deploy the application to the cloud and trace its model and tool calls.
  7. 07Track quality, latency and cost, and handle failures safely with guardrails.

02 / THE FINAL PROJECT

You finish with something real.

You build and deploy an AI assistant for a concrete company use case, for example one that answers questions from messy internal documents with sources and carries out actions after human approval. You show where it works, where it fails, and how you measured that.

Alexey Grigorev

03 / WHO TEACHES

One instructor. Every session, every project review.

I'm Alexey Grigorev. I founded DataTalks.Club and created its free Zoomcamps: Data Engineering, Machine Learning, MLOps, LLM and AI Dev Tools. Their course repositories have more than 85,000 stars on GitHub. I wrote Machine Learning Bookcamp, and I teach paid cohort courses on both of these topics on Maven.

This program is different from the Zoomcamps in one way that matters: it is small. I teach every day myself and I look at every participant's project every week. There are no teaching assistants and no pre-recorded lectures in place of a class.

04 / WHO IT IS FOR

This is for you if

  • You program today, ideally in Python, and can use APIs and work in a repository.
  • You can study full-time, Monday to Friday, for the whole program.
  • You can follow technical teaching and work with others in English.
  • You want your next job to be in software or AI engineering, including if you are between jobs right now.

This is not for you if

  • You have never programmed. Start with the free Zoomcamps and come back.
  • You need an evenings-only course.
  • You want a certificate without building anything.

05 / WHAT WE COVER

Sixteen modules, one project that grows every week.

This is the draft syllabus. The shortlist's survey answers shape the final version.

Weeks 1–4Retrieval
  1. 01A Python service on an LLM API.You produce: a small API that returns validated, structured answers.
  2. 02First assistant with retrieval.You produce: a knowledge assistant that answers with sources.
  3. 03Document processing, ingestion and indexing.You produce: an ingestion pipeline you can rebuild with one command.
  4. 04Vector and hybrid search.You produce: three retrieval variants, with examples of where each fails.
Weeks 5–8Agents
  1. 05Evaluating retrieval.You produce: a retrieval evaluation and a justified configuration.
  2. 06Tools, function calling and the agent loop.You produce: an assistant that calls an API through a tool.
  3. 07Agent frameworks, MCP and context.You produce: the assistant rebuilt on a framework, with one MCP tool.
  4. 08Serving and cloud deployment.You produce: a deployed assistant with cited answers and one tool (midterm).
Weeks 9–12Quality
  1. 09Agentic workflows and human approval.You produce: a workflow action that waits for approval and handles refusal.
  2. 10Testing LLM applications and CI/CD.You produce: a regression suite running in CI.
  3. 11Evaluating answers and agents.You produce: an evaluation report comparing two configurations.
  4. 12Tracing, observability and monitoring.You produce: traces and a monitoring dashboard for the deployed assistant.
Weeks 13–16Production and capstone
  1. 13Cost, latency and model choice.You produce: a cost and latency report with one measured improvement.
  2. 14Safety: guardrails, injection and compliance.You produce: tested guardrails and a short compliance checklist.
  3. 15Capstone: build and measure.You produce: the capstone build with evaluation evidence.
  4. 16Capstone: improve, review and defend.You produce: a deployed assistant, an evaluation report and a defence.

06 / FUNDING

A Bildungsgutschein pays for a seat. It doesn't get you one.

I'm applying for approval so that the Agentur für Arbeit can cover the tuition with a Bildungsgutschein. That approval is still pending, and the Agentur decides on each person's voucher individually. Nobody can promise you funding today, and I won't.

Funding and admission are two separate decisions. The Agentur decides whether it pays. I decide who is in the cohort. You need both.

If you pay yourself, the full program will cost €9,000 to €10,000.

How the Bildungsgutschein works (arbeitsagentur.de)

07 / HOW YOU GET IN

Why 15 seats

I teach this cohort myself and review every project. I can do that properly for about 15 people, not for more. In Berlin the room doesn't fit more either. If more people qualify than there are seats, there is a waitlist and a later cohort. I won't add seats.

  1. 01

    Join the shortlist. Leave your email on this page. It takes ten seconds and commits you to nothing.

  2. 02

    Apply first. When applications open, the shortlist gets them before anyone else.

  3. 03

    Short task and a conversation. A small practical exercise and a 20-minute call, so we both know the level fits.

  4. 04

    Seats are offered in rounds. Everyone who applies gets an answer.

08 / QUESTIONS

Questions

Is this program confirmed?

It is planned for April 2027. Two tracks are on the table, agentic engineering and AI engineering, and I'll run the one with more serious interest first. Joining a shortlist is how you vote.

What do I get for joining the shortlist?

You get the application before it opens publicly, and updates on dates, format and funding approval. Nothing else, and no obligation.

What do I need to know before I start?

You are comfortable programming, preferably in Python, and you can use the command line, Git and basic Docker. You don't need a machine-learning background, advanced mathematics or a GPU.

Will the Agentur für Arbeit pay for it?

That is the plan. I'm applying for the approval that makes Bildungsgutschein funding possible, and it has not been granted yet. Even after approval, the Agentur decides each person's case. Talk to your adviser early.

What does it cost if I pay myself?

€9,000 to €10,000 for the full program.

Is it online or in Berlin?

Not decided yet. It will be either live online from Germany or in person in Berlin, in Prenzlauer Berg near S+U Schönhauser Allee. The survey after you join asks which one you could attend.

What is the schedule?

Full-time: Monday to Friday, 09:00 to 16:00 Berlin time, live, for 12 to 16 weeks. The exact length is fixed together with the approval. I'm also considering a part-time version; tell me in the survey if you need one.

Do I have to live in Germany?

The program is built for people in Germany, and a Bildungsgutschein is only available there.

How do you choose who gets a seat?

By the same published criteria for everyone: programming experience that fits the track, availability for the full schedule, English, a concrete goal, and the practical task. Your employment or benefit status is not a criterion.

What happens to my email?

I use it only to write to you about this program and its application. You can ask me to delete it at any time. Details are in the privacy notice.

09 / JOIN THE SHORTLIST

Fifteen seats. The shortlist hears first.

Leave your email and I'll write to you before applications open.

15 SEATS · 1 INSTRUCTOR · NONE OFFERED YET

15 seats · April 2027