01 · Education, presentations, workshops
AI Under the Hood
Enterprise AI training that explains the technology, the economics, the risks and the rules, from the KV cache to the bottom line. Built from eight years of AI work and thirty in IT.
Every figure on the slides has its source and its date. Every demo is measured on our own hardware. Two minutes about Masbando at the very end, nothing in between.
Formats
From 75 minutes to two days.
The same slide repository at different depths: the briefings are the full training cut to its essentials, so what a manager hears and what an engineer hears fit together.
Two days
AI Under the Hood: the full training
Thirteen sections, from how a language model works to the business case, the law and the organisation. For IT, architecture, security and data teams, and for leaders who want the whole picture.
- On your premises or in Zurich
- 6 to 15 participants, one company per course
- Workshop: each participant works on a use case of their own
90 minutes
AI Under the Hood: the IT briefing
How LLMs actually work (tokens, context, the KV cache), when to run a model locally and when to use an API, the security picture with prompt injection shown live, and why one gateway in front of all models pays back.
- For IT managers, architects, security, platform and data teams
- Two live demos, measured on our own hardware
- One page to take home: the decision framework and the first 100 days
75 minutes
AI and the Bottom Line: the management briefing
Where AI pays off, what it really costs, how to keep it under control, what it does to people and skills, and how to pick a first use case. No technical background needed.
- For heads of department and team leads
- Each participant scores one use case of their own on a canvas
- You leave with a ranked shortlist and the structure of a business case
Half a day
Workshops and keynotes
A half-day workshop on your own use cases: prioritisation, the business case, the first 100 days. Keynotes for management meetings and events, cut from the same material.
- Agenda agreed with you in advance
- Slides in English, delivery in German or English
Curriculum
Thirteen sections.
The full training covers all thirteen. A briefing takes the essential slides of the sections that matter to its audience, so the depth varies while the content stays consistent.
Every section ends with four questions to check your understanding, and a glossary closes the deck.
- 01
AI Foundations and Landscape
What AI is and is not, the history, training data, AI in business, model families, LLMs versus classic machine learning, the market and its impact, limits and outlook.
- 02
How LLMs Work
Tokens and why German costs more, the context window, prompt and context, prefill and generation, capabilities and limits.
- 03
Under the Hood
The transformer, the KV cache, dense versus mixture of experts, diffusion, quantization, training and adaptation, distillation, interpretability, watermarking.
- 04
Measuring LLM Performance
Speed metrics and how to benchmark them, intelligence benchmarks, your own evals, behaviour, cost and efficiency.
- 05
Building with LLMs
Prompting, retrieval (RAG), tools and MCP, agents and the harness, skills, memory, the API call, coding assistants, voice agents, fine-tuning, use cases.
- 06
Local or API?
Hosting options, the decision rule, open models, hardware, the inference stack, clustering, reference architectures, operations.
- 07
Security and Risk
The threat landscape, prompt injection shown live, AI-enabled attacks, deepfake fraud, shadow AI, safeguards in the right order.
- 08
The Bottom Line
Token economics, the total cost of local hosting, the business case, cost control, make or buy.
- 09
Organisational Aspects
Strategy, use-case selection, operating model, governance and policy, people and change, maturity and roadmap.
- 10
Legal and Regulatory
Data protection, data sovereignty, the EU AI Act, IP and copyright, secrecy and confidentiality, liability and contracts, sector rules, a compliance checklist.
- 11
The AI Gateway
Why a gateway, what it pays back, a reference design, trust by design, running it, choosing and introducing one.
- 12
Current Trends in AI
A trend radar: efficient models, agentic trends and the agentic web, open and sovereign AI, physical AI, and what they imply for the next twelve months.
- 13
Demos and Labs
Live demos measured on the machine in front of you, and a workshop in which each participant works on a use case from their own unit.
Measured
Measured, not quoted.
Three figures from the IT briefing, all our own measurements. Vendor claims are shown as such; what we can measure ourselves, we do, and the reference results travel with the deck in case the network fails.
- 50 of 50
- runs followed the hidden instructionA small model summarising a supplier e-mail passed the attacker's request on every time; a warning in the system prompt changed nothing.
- 26×
- faster first token with the cache keptA question after an 8,000-token document: 1.42 seconds, or 0.055 with the document's KV cache kept in memory.
- 45×
- cheaper with the cost leversTen questions about one 50-page contract: from 5.43 dollars to 12 cents at list prices, same questions.
Outcomes
What you take away.
An in-depth understanding of AI and what it means in the commercial environment. Each goal is an outcome; the line below it is the knowledge that gets you there.
Use AI more efficiently, with better results
Understand how LLMs work, deep enough to judge vendor claims and read an API bill.
Reduce costs
Know what drives them, which levers exist, and how to build a business case with a baseline.
Keep data secure
Know the security guard rails, and when to host locally or use an API.
Develop AI governance
Know the legal guard rails and how to build an organisational roadmap.
- Materials
- Participants get the slides afterwards as a view-only link and as a PDF. Sources for any figure on request.
- Demo setup
- A demo machine with a GPU travels with the trainer; everything shown live is measured on it.
- Trainer
- Alexander von Malmborg, MSc EE ETH Zurich · MBA INSEAD. Thirty years in IT, eight in AI. Founded Masbando GmbH in 2009.
- Fees
- On request.
Contact
Book a briefing.
Reply with two or three dates that suit your team and a room. We come to you, in German or English.
- Phone
- +41 44 244 90 40
- info@masbando.com
- Office
- Masbando GmbH, Zurich