Skip to content
Felix Schumann
Get in touch

Services / Prompt engineering

Tell the AIwhat you actually need.

Good AI results start with a clear task, the right information and a verifiable output format. I develop prompts and context structures for specific workflows and test them against examples from real use.

See it builtDiscuss your project

Prompt engineering · Context, data, review points

Explore an example

Task + context
Task01

What should the result look like?

  • 01Goal and audience
  • 02Expected format
  • 03A concrete example

Your task shapes the implementation.

The everyday problem

A useful answer should not be a lucky accident.

A prompt works once and fails on the next example. Instead of adding ever longer instructions, I separate the task, required knowledge and expected result. We test variants against the same cases and see which ones help in the actual workflow.

A successful one-off chat becomes a reusable way of working. Roles, examples, sources and limits are clearly described. Variants are compared on the same tasks so improvements and the need for human review become visible.

What we will work on together.

The scope follows your project.
  1. 01

    A clear template

    System prompts, templates and defined output formats

  2. 02

    Compare results consistently

    Context preparation, examples and verifiable quality criteria

  3. 03

    Use it in your workflow

    Integration into the appropriate assistant or workflow

From my work

See what this looks like in practice.

All projects
In daily use2026

Content intelligence and production platform

A custom algorithm. Clear plans. Your content agent.

A custom-built algorithm, an editorial planner and an agent tailored to your content goals. From research to production.

  • A custom algorithm for your topics
  • A planning tool for the entire workflow
  • A content agent that fits your strategy
  • Next.js
  • TypeScript
  • MongoDB
  • +1
Explore the projectTwo editorial briefings per day
Shipped2026

Context preparation for AI development

Give the coding agent the context that counts.

Relevant code and compact logs keep AI inputs focused. Local checks help detect invalid edits early.

  • Select relevant code sections
  • Validate edits before writing
  • Record technical decisions
  • Python
  • LM Studio
  • Qwen
  • +1
Explore the projectClient project

Our first conversation

Bring one useful answer and one that missed the mark.

You do not need a finished technical specification. Your idea, an example or a look at the current workflow is enough to start.

Let us talk it through
  1. 01

    What separates a useful result from one that merely sounds good?

  2. 02

    Which examples, sources and boundaries should the model consider?

You might also be wondering:

Which services do you offer?How do you handle data and external AI services?How do you estimate the scope and prepare an offer?

Inside the implementation

MCP integration: Useful agents need understandable toolsTool names, inputs and outcomes: design an MCP integration that AI agents can use predictably and people can verify.Context engineering: The right information at the right momentWhat an agent needs to know now, what can be loaded later and why a longer context does not automatically produce better work.RAG or fine-tuning: Which problem are you trying to solve?Retrieve current knowledge or improve repeated behavior: a practical decision guide and evaluation plan for internal AI assistants.From AI text to dependable data: Using structured outputs wellValid JSON does not guarantee correct facts. A practical design for extraction, business validation and understandable failure handling.
Felix SchumannFelix SchumannAI automation engineer

Name it, I build it.

If something here looks like the problem you have, write to me. A short description of what is not working is enough to start.

Get in touch[email protected]