Your best people
wrangle data
half the week.
Converting, projecting, tiling, classifying, repairing broken metadata. The same pipeline for the fortieth time. Those hours either come off your margin or crowd out the work you would rather be selling.
AI Agents can do a lot of that now - driving PDAL, GDAL, Python, and the rest of your stack for you. So your team can focus on the hard stuff.
In two half-days I show you and your team how, on your own data, using widely available tools.
- Duration
- 2 half-days
- Group
- Up to 6
- Where
- Your office or remote
- Language
- German or English
You type this:
“Here's a lidar scan of a glacier, four epochs, all different file formats. Calculate the volume change across all epochs.”
The agent does this:
- Converts, merges, and trims the data
- Calculates the volume change across all epochs
- Validates the results and writes a ready report.
The only thing you need is a Claude license, and the tools your team already runs.
The two
half-days
Hands-on in the second half-day, after a solid foundation in the first. So you walk away with real results right away.
Getting agents to do real work
1. The mindset shift
An agent with a terminal, not a coding tool. “Here's a folder of LAZ files, ground-classify them and give me a quality report.”
2. Wrangling and conversion
Batch conversion across E57, LAS/LAZ, COPC, Shapefile, GPKG, GeoParquet. Reprojection, tiling, merging, repairing broken CRS metadata.
3. Point cloud pre-processing
PDAL pipelines described in plain language: denoising, thinning, clipping, ground classification, DEM and DSM generation.
From one-off to something you keep using
4. Analysis and reporting
Volume calculations, change detection, zonal statistics, plots, a formatted report at the end. Dataset to deliverable in one session.
5. Your first agent
Turning a workflow into a reusable skill or script. Running it headless on a schedule or a watch folder, with checkpoints and validation.
6. Trust
Where agents silently get geo wrong: CRS, units, nodata. The module that decides whether any of this reaches a client deliverable.
Plus
The compliance questions you'll be asked internally anyway: data privacy, on-premise and EU hosting, what to tell your public sector clients.
You keep
what we build.
Working automations for two or three of your own recurring chores, built on your data during the workshop.
A project template so the next automation starts from something instead of nothing.
A verification routine for the places an agent gets geo quietly wrong. Your real risk is a wrong number in a deliverable.
Materials and 90 days of async (email) support.
Alex Key,
not a trainer.
I build AI agents for spatial data as my actual work - for the companies that make the software you use.
Geoinformatics background, fifteen years shipping software, and before AI I led software and product teams at BMW, Audi and DriveNow and ran my own engineering studio. I am not teaching you a tool I read about.
The part that matters for this workshop: I know where agents get geo wrong, because I spend my weeks finding out. That is a different thing from a general AI trainer who has never opened a PDAL pipeline, and it is why module 6 exists.
Short videos of agents doing real spatial work - glacier change detection, a CloudCompare integration, facade analysis.
One price,
no seat math.
Net, flat, for the whole team. Not per seat.
- Two half-days, up to 6 people. €500 per additional participant, to a hard cap of 8. Past that it stops being hands-on.
- On-site, travel within Germany, Austria and Switzerland included.
- One company at a time, so the workshop is built around your data rather than shared with strangers.
- If the workshop doesn't deliver, I keep working with your team until it does. A follow-up session, at no cost.
Questions people
ask first.
What does the team need beforehand?
Laptops they have admin rights on, and a Claude subscription per participant. I send exact setup instructions two weeks ahead and help if anything fights back. Nobody needs to install anything clever.
Does anyone need to be able to program?
No. It helps if some of them have written a script or a model builder chain at some point, because they will recognize what the agent is doing. It is not a requirement.
On-site or remote?
Both work, same price - travel is included either way. I would push for on-site: it is noticeably better when someone gets stuck.
What happens after the workshop?
90 days of async questions and support included, usually via email or video calls if needed.
Is this going to be out of date in six months?
The specific commands, partly - the tools move fast. The way of working, no. And that is what the workshop is actually about.
Two half-days.
Then your team keeps going.
If this looks useful, reply to me on LinkedIn and we'll work out whether it fits and when.