A skill record you can trust
Not minutes watched. Every verified skill links to the graded work behind it: tests passed, rubric scores and the evidence Claude quoted.
The LLM Engineer masterclass
Proof of skill.
Not proof of attendance.
Your engineers learn LLM engineering by building real systems. Claude grades every exercise with evidence, practice follows them into Teams, Slack and WhatsApp, and you get a skill record you can trust.
- Real builds
- Graded with evidence
- Hosted in the EU
LLM Engineer › 5.2 RAG
Modules
- 4 Inside the Model
- 5.1 LLM APIs
- 5.2 RAG
- 5.3 Agents
- 7 Adapting Models
- 8 Evals
▲ 3 skills verified
LearnQuizPracticeRole play
Embeddings, vector search and RAG
Graded by Claude84/100
Retrieval logic is correct and tested. Add a reranker to fix the off-topic chunks in test 4.
def retrieve(query, k=5):
vec = embed(query)
hits = store.search(vec, k)
return rerank(query, hits)
▶ RunSubmit
✓ 4/5 visible tests
Tutor
Why do you think test 4 returns chunks about billing for a question about refunds?
The problem
Completion is not competence.
Traditional platforms host videos and count who finished them. Nobody can tell you who can actually ship an AI system, and the EU now expects you to show that your people are trained.
- 20 to 30%
- Typical completion for voluntary courses on a traditional LMS.
- 0
- Things a completion certificate proves about what someone can actually build.
- Aug 2026
- When national authorities could start enforcing the EU AI Act's AI literacy duty.
How it works
Every lesson ends in work you can show.
Four steps per lesson, unlocked in order. Each module ends with a real build that has to pass hidden tests before the next one opens.
- 01
Learn
Short, dense lessons with diagrams, math and runnable code. No hour-long videos to sit through.
- 02
Quiz
A quick check of the ideas. 80% to pass, with shuffled questions on every retry.
- 03
Practice
Write real code in the browser. Hidden tests run on Submit, then Claude grades it against the rubric, quoting your code as evidence.
- 04
Role play
Defend your design to a skeptical PM, a security reviewer or a CFO played by AI. Graded on what you actually said.
AI tutor in the editor
It sees your code and failing tests, and asks the question that gets you unstuck. It never hands over the answer.
Placement test first
Already fluent in Python? Skip those modules and start where the LLM work begins.
In your tools
Learning that follows them into work.
Between lessons, a learning agent sends recall questions, quick quizzes and short role plays in the tools your team already has open.
Microsoft TeamsSlackWhatsApp
Recall that sticks
Short questions spaced over days, aimed at the skills each person is weakest on.
On their schedule
Learners pick their channel and their time windows. No random pings during a deploy.
Counts everywhere
Every answer, in any channel, feeds the same verified skill record.
# learn-samRecall · 2 min
BuildStackHQ Learn 10:42
Your RAG assistant cites the wrong document about 1 in 10 times. What do you check first: chunk size, the embedding model, or a reranker?
Sam 10:44
Reranker. Retrieval finds the right chunks in the top 20, they just are not ranked first.
BuildStackHQ Learn 10:44
Correct. Recall at 20 is fine, precision at 1 is the problem.✓ Retrieval tuning strengthened
The first masterclass
LLM Engineer: beginner to master.
From tokens and attention to RAG, agents, fine-tuning, evals and production. Every module ends with a system you would actually ship, and the capstone is built on your own company's data.
- modules
- 10
- modules
- lessons
- 33
- lessons
- real builds
- 10
- real builds
- capstone on your data
- 1
- capstone on your data
- 017 lessons
Python Foundations
Build: Document intake CLI
- 023 lessons
Machine Learning Foundations
Build: Ticket classifier from scratch
- 035 lessons
Python Intermediate
Build: LLM client library
- 046 lessons
Inside the Model
Build: Domain language model
- 053 lessons
Building with LLMs
Build: Internal knowledge assistant
- 063 lessons
Python Advanced
Build: Production RAG service
- 072 lessons
Adapting Models
Build: Cost-cutting fine-tune
- 081 lesson
Evals
Build: Eval platform in CI
- 092 lessons
Production
Build: Production hardening
- 101 lesson
Frontier and Mastery
Build: Frontier technique applied
For engineering leaders
Know who can ship, not who clicked through.
Everything a learner does turns into evidence you can act on, and the same record covers your AI literacy obligations.
AI Act training evidence
Article 4 asks companies to show documented, ongoing AI literacy effort. Export the record per team whenever you need it.
Readiness and ROI
See which teams are ready to ship AI systems, where they are stuck, and what the capstones built on your own data.
Your data stays in the EU
Hosted in Frankfurt. Claude runs on Amazon Bedrock in the EU, so learner work and company data stay inside the EU.
Pricing
One simple price per seat.
Everything included. GPU time for the fine-tuning modules is an optional add-on.
Monthly
€49per seat / month
Cancel any time.
- Billed monthly per active seat
- Add or remove seats any time
Best value
Yearly
€490per seat / year
Two months free.
- Billed once a year
- Seats can be reassigned between engineers
Every seat includes
- The full LLM Engineer masterclass, 10 modules and capstone
- Placement test that skips what you already know
- Browser coding with hidden tests and Claude grading
- AI tutor and graded role plays
- Practice in Teams, Slack or WhatsApp
- Verified skill record and certificate
- Team dashboard and AI Act evidence export
FAQ
Questions teams ask first.
Who is the masterclass for?
Software engineers who want to build, evaluate and run LLM systems in production. It starts from Python basics, but the placement test lets experienced engineers skip straight to the LLM work.
How does AI grading stay fair?
Every grade follows a published rubric, and each point comes with the evidence it was based on. Automated tests run first, scores are recalculated by the platform rather than trusted from the model, and learners can always resubmit.
Do learners have to use WhatsApp?
No. The company decides which channels are allowed, and each learner picks Teams, Slack, WhatsApp or none. WhatsApp is always an explicit opt-in.
What about the fine-tuning modules?
Fine-tuning needs GPUs. They are available as an add-on, or learners can connect their own GPU account.
Does this make us AI Act compliant?
It gives you documented evidence of AI literacy training for your engineers, which Article 4 asks for. It is not legal advice and does not cover your other AI Act obligations.
Turn your engineers into LLM engineers.
We are onboarding the first teams now. Tell us how many engineers you want to train and we will set you up.