Graculus

Personal training in instructing AI. Where do you stand?

  1. Hatch

    After this level You understand how to instruct AI, and you reach the limits of what AI can do for you within a single session.

    • What AI is and how AI works
    • Hands-on experience with AI-assisted deciphering of notes, proofreading, formatting documents, having technical concepts explained, troubleshooting, translation, advanced searches, implied scripting, scouring online vendors, and more
    • Image and video generation: exploring styles and partial image regeneration
    • AI voice generation
    • Browser-driving agents
    • When to use which model
    • Sessions and compaction

  2. Walk

    After this level A single session no longer limits you. Without programming, you build AI projects that persist across any number of sessions.

    • Programs without programming: websites, scripts, installed programs, server applications
    • Writing project instructions that AI follows
    • Basic automated AI execution
    • What memory is, and why to avoid it
    • Pitfalls of basic orchestration
    • Secrets management

    Work at this level Waymarked, a website for a themed trail with stations: map, GPS, texts and audio on one screen. At this level, a first working version takes a few days. You build it yourself, so you are next to the people using it: what they stumble over reaches you directly, not through a chain of contractors.

  3. Fly

    After this level Growth no longer limits your projects: they stay maintainable as they get larger. You no longer just use your AI environment, you make it.

    • Technical debt: how it affects AI projects, why it is not just about code, and how to fight it
    • Applications that work: design, build, test, deploy
    • Writing your own skills and custom agents
    • How and why to rewrite the AI's system harness
    • Rule files and hooks
    • Advanced automation: scripted execution of AI tasks

    Work at this level ClassyQuiz, a quiz site with no AI built in. The visitor's own AI reads the site's instructions and builds the quiz.

  4. Soar

    At this level The vendor's tools no longer limit you. You call AI from your own programs and have AI drive AI, safely and without unwelcome surprises.

    • Automation beyond the vendor's tools: direct API use, DSPy, OpenClaw
    • Injection defence
    • Advanced delegation and subdelegation
    • Effective memory
    • Self-evolving skills

Personal training

One to one, since no two people start from the same place

Pace
Yours: several hours in the first week to get going, an hour a week, or a session every few weeks when new questions have come up.
Material
Your own work, in the tools you or your employer already use. Where those can't be used for learning, I provide an environment.
Prerequisites
No prior knowledge needed, only curiosity.
Where
In person in the Baden and Zurich area; online where meeting isn't practical.
Language
English or German.
Confidentiality
What you show me stays between us, in writing if you or your employer want that.
Teams
On request I organize workshops at the Hatch level. Above that level, people's speed and prior knowledge often drift too far apart for an effective group workshop.

Price

One hour
CHF 250
Ten hours
CHF 2,200
Your first hour
CHF 220

If during your first three hours you are not satisfied with the quality of training, say so and you get your money back.*

* Tell me in the session, at the latest at the end of your third hour. We stop the training there, and I refund everything you have paid, including hours paid in advance.

Write or call

peter@graculus.ch
+41 00 000 00 00

Two lines are enough: what you do with AI today, and what you would like to be able to do. You get an answer within one working day, with a proposal for a first hour.

Who teaches

Photo

Peter Bänziger. Twenty years in software, most of it on high-frequency trading systems. I have worked as developer, project manager, Scrum Master and release manager, set up build and test pipelines, and brought old codebases back under control.

I have sat on the engineer's side of the table, the manager's and the user's, which helps when teaching all three. I teach what I do the rest of the week.

An Alpine chough on a summit

The rest of the week

  • Software, built with AI

    From the idea to an application that is deployed and maintained.

  • Agent configuration

    Skills, agents, instructions and unattended runs, set up in your team's environment.

  • Integrations

    AI connected to the systems and data you already have.

If you have a project rather than a wish to learn, I take it on as Peter Bänziger Software Consulting. Write to me about it.

Selected open source work

Three recent projects of my own, each built with AI and each a different kind of work.

  • Waymarked

    Interface design, tested with its users

    A website for a themed trail: map, GPS, texts and audio for each station, on one screen, with no break when moving from station to station, to the map and back.

  • ClassyQuiz

    Instructions for the visitor's own AI

    A quiz site with no AI built in. You point your own AI at the site; it reads the site's instructions and hands back a quiz to play alone, with friends or with a class.

  • NTRIP mountpoint map

    Data research by unattended agents

    A world map of free GPS correction services. Agents research the data unattended overnight, and the map rebuilds itself four times a day.