AI-CONTEXT · Path 1: Understand and use safely
Prompt and Context Engineering
Two days after which good AI results in your team are no longer a matter of luck, but repeatable building blocks.
- Duration
- 2 days
- Group
- up to 12 people
- Format
- In-house · Remote · Open enrollment
- Language
- German or English
- Price
- €15,800 flat, plus VAT
Who this track is for
Heavy users, subject-matter experts and developers who work with AI daily and whose results are reused by others: analyses, reports, data extraction, text modules, precursors to automation. Also for teams that finally want to share their best prompts instead of burying them in private chats.
Who it is not for
Not for beginners who do not yet use their own prompts in daily work, they start in AI-LIT-LAB or AI-WORK. Anyone who wants to connect knowledge bases systemically belongs in Enterprise RAG Engineering (AI-RAG).
Starting situation
A few people in the organization achieve impressive results with AI but cannot explain why, and nobody can reproduce them. Prompts live in private chat histories, outputs vary from run to run, and as soon as a result needs to go into another system, manual work begins. The team lacks patterns, structure and a shared repertoire.
This exists after the track
- A set of reusable context patterns: roles, domain context, examples and limits as building blocks that can be combined per task
- Prompts with structured outputs: JSON and tables that can be taken straight into Excel, further processing or downstream systems
- Quality and hallucination controls per prompt: test cases, edge cases and defined checks before reuse
- A team prompt library version 1: named, documented, tested, with maintenance responsibility clarified
- Two documented training days as a building block of your Art. 4 competence evidence
Prerequisites
Regular hands-on AI use in daily work is assumed, roughly at the level reached after AI-WORK. The JSON exercises require no programming skills, developers get additional depth from the same exercises.
Preparation before the track
Each participant submits their three most-used prompts in advance, untidy is fine, plus one example each of a good and a useless result. These real prompts are the raw material for both days.
What's included
- 2 training days with Dino Bordonaro, in-house or remote
- Advance evaluation of the submitted prompts as the team's starting picture
- Pattern catalogue of context patterns and output schemas as editable templates
- Prompt library template including naming convention, versioning and test case structure
- Attendance certificates and audit-proof training documentation
- 45-minute remote office hour four weeks later on evolving the library
Agenda
Day 1
Why the same prompt answers differently ten times
The submitted prompts under stress test: the same prompt, multiple runs, measured variance. This makes visible which formulations carry weight and which are left to chance. Each participant picks their most important prompt as their working object for both days.
Context patterns instead of gut feeling
Patterns are distilled from the best submissions: role, domain context, examples, limits, negative instructions. Exercise: each participant dissects their own prompt into building blocks and names which block has which effect, verified by leaving blocks out.
Context from company knowledge
How much document, table or background knowledge belongs in the context, and in what form? Exercise on real material: feed the same document as raw text, as an excerpt and as structured bullet points into the context and compare result quality. Plus the data rule question: What is allowed in at all?
Day result: your pattern set passes the test
Each participant rebuilds their working prompt fully on context patterns and runs it against three predefined test cases. It passes if all three cases produce usable results that are demonstrably better than the original version. The patterns go into the shared catalogue.
Day 2
Structured outputs: JSON and tables
From prose to reusable results: define output schemas, enforce mandatory fields, format tables cleanly for Excel. Exercise: a data extraction from unstructured documents whose result drops into a table without rework. Developers additionally validate the JSON by machine.
Quality and hallucination controls
Every library prompt gets test cases and edge cases: empty inputs, contradictory information, missing data. Exercise on prepared documents with planted errors: Which controls catch the invented number, which let it through? This produces the review routine per prompt.
The team's prompt library
The individual results become a shared repertoire: naming convention, description, usage limits, test cases, version. Workshop: the team's ten most important prompts are made library-ready and cross-tested. Team decision: Where does the library live, who maintains it, how do new prompts get in?
Day result: library version 1 passes the acceptance test
The real test: each participant solves someone else's task using only the other participants' library prompts. Whatever works without questions is accepted, the rest gets concrete rework with an owner and a deadline. Closing: maintenance plan and the first three expansion candidates.
Exercises and lab share
Both days are workshop work on the submitted real prompts: stress tests with multiple runs, rebuilding on patterns, JSON extraction from real documents, and a mutual acceptance test of the library.
Platforms
In-house we work with the AI tool approved in your organization, in whichever version is available to you. On request we provide access on our lab environment, in which case work is done with anonymized or practice documents.
Transfer evidence
Two verifiable day results: each participant's pattern set passes three defined test cases, and the team library passes the mutual acceptance test. Both results are documented in an audit-proof manner.
Artifacts you take home
- Personal context pattern set with proven improvement over the original prompts
- Output schemas for JSON and tables as editable templates
- Test case and review routine template per prompt against hallucinations
- Team prompt library version 1 with maintenance plan
- Training documentation for the compliance archive
Optional extensions
- Enterprise RAG Engineering (AI-RAG) when company knowledge should enter the context systemically instead of by hand
- Working Safely and Productively with AI (AI-WORK) for colleagues not yet at heavy-user level
- The company's Sovereign Assistant when the library should run on an on-prem environment
Boundaries
This course turns your team into confident prompt authors with their own library. It replaces neither a RAG implementation, nor application development, nor platform operations.
Frequently asked questions
Will better models soon make prompt engineering obsolete?
Better models are more forgiving of sloppy prompts, but repeatable quality, structured outputs and tested controls remain craftsmanship. That is exactly why context is in the course name: whoever masters context and test cases benefits from every model upgrade instead of starting over.
Are there open dates with individual seats?
No, we deliberately deliver this track in-house: €15,800 plus VAT per delivery with 4 to 12 people. That way everyone works on your organization's prompts and documents instead of third-party examples, and the shared prompt library stays entirely with you.
Do participants need programming skills for JSON?
No. JSON is introduced as a structuring format and developed in the exercises with templates, which business users manage routinely. Developers in the same course deepen machine validation on the same exercises, which benefits the shared library.