Alex Key
Independent AI engineer · Germany
Agents that take over the work around your point cloud processing - the setup, the re-runs, the checks, the deliverables - so your team gets its hours back, with results held to your own quality standards.
I work hands-on, as a contractor, with your tools and your team - from finding the right workflow to an agent in production.
The hours around the actual processing are priced in as unavoidable. They are not, anymore:
Project setup and moving data between tools - formats, coordinate systems, folders, naming.
Running the steps, checking the result, running again with different parameters, fixing the last five percent.
Assembling deliverables and reports - the part nobody was hired for and everybody does.
Agents that execute long-running manual tasks on their own, while keeping humans in full control. Not a chatbot that talks about your data, not a model that detects a little better - an agent that opens your tools, runs the steps, checks the results and brings you the deliverable.
Nobody believes this from a paragraph. Watch one run.
Videos are hosted by Vimeo and remain blocked until you explicitly load one. Details are available in our privacy policy.
Agents don't chat about your survey data. This one runs a complete analysis on its own: two epochs of an alpine glacier in, a change result out - a multi-step task carried end to end, unattended.
Long-running, multi-step work on the data itself. That is the difference to a chat window.
Spatial data is more than point clouds. This agent runs Monday triage across a portfolio of fields: it reads crop stress from satellite imagery, reasons about the likely cause field by field, and plans which fields an advisor should visit first.
The judgment work sits right next to the data work - the agent does not stop at the number.
Agents are most useful when they can reach for the tools experts already use. Here an agent drives CloudCompare to run a volume calculation for slope monitoring - the same software, the same steps, no hands on the mouse.
No new platform. Your existing software, driven by an agent.
Pairing graph-based point-cloud segmentation with an LLM agent that reasons over a scene and edits it to spec. Presented at the XXV ISPRS Congress 2026 in Toronto and published in the ISPRS Archives.
This is not an IT project. Projects turn around faster without adding headcount, your experts spend their time on judgment instead of clicking through the same pipeline again, and the work stops depending on the one person who knows every tool's quirks.
We look at where the hours actually go and pick the chore where an agent saves the most - not where the hype points.
The agent works in the tools you already license and trust - point cloud software, GIS, scripts, spreadsheets. Nothing to migrate, nothing to re-learn.
Documented, observable, and owned by your team when I leave - not a black box that needs me to babysit it.
Best fit: laser scanning, reality capture and survey teams that run point cloud processing at volume - several people on the pipeline, projects every week. If expert hours are going into repetitive work, that is usually where an agent pays for itself first. I take on a small number of engagements at a time.
You pick one of your datasets and one chore. I build an agent that automates it, on your data, and show you where it holds up and where it doesn't. The fastest way to find out what agentic automation looks like for your team.
Hands-on: I build the agent into your workflow, with your tools, measured against your quality standards - and leave it running in your team's hands.
Two half-days with your team on working with Claude Code on geo data. The smallest first step - your people automate their own chores from day one.
I also help spatial software vendors bring AI agents into their products - so the workflows your users run by hand today become something your product does for them, with reliability you can measure.
Talk to me about your product
I'm an independent AI engineer based in Germany. I work hands-on - from rapid prototyping to production-oriented system design.
My focus is LLM agents and orchestration for spatial and 3D data, and the evaluation that keeps them reliable. I studied geoinformatics and mining geomatics. Before AI, I led software and product teams at BMW, Audi, and DriveNow, and ran my own engineering studio, Kawunu, delivering production systems for clients in mobility, manufacturing, healthcare, and logistics.
For the longer version, see my background on LinkedIn.
