Terrain AI — Bridging Geoinformatics & AI

by Dr. Ferdinand Vettermann

From spatial intelligence to production AI systems — building at the intersection of geoinformatics, machine learning, and tech leadership.

About Me

I am a Data Scientist and AI team lead with a core background in geoinformatics. My work operates at the strict intersection of geospatial science, modern machine learning, and software engineering — scaling solutions from high-resolution satellite image segmentation to agentic AI systems.

With a PhD from the University of Rostock and over a decade of experience spanning academic research, applied data science, and technical leadership, I combine deep domain expertise with production engineering execution. I build systems that transform raw spatial data into actionable, automated intelligence.

Featured Projects

Exploring the terrain where spatial intelligence meets modern AI

Live

Terrain AI: Portfolio Architecture

A zero-framework, high-performance technical showcase. Optimized for sub-second loading times, perfect Lighthouse scores, and a sophisticated geospatial aesthetic.

Vanilla JS HTML5/CSS3 Web Performance UX Strategy
In Progress

Ants on Earth — Part 1

Multi-agent reinforcement learning with an ant colony navigating real-world elevation data. Agents emerge collective pathfinding behaviour across actual terrain relief.

Reinforcement Learning Python Multi-Agent Systems Geospatial
Done

Jobhunt

Semantic matching engine for unstructured job data with spatial constraints. Uses LLM-driven pipelines to parse, embed, and rank results against geographic and domain-specific criteria.

LLM Python API Semantic Matching Containerization AI assisted developing
Done

Political Cartography

Anamorphic mapping of electoral data to visualize density-weighted trends. Transforms geographic boundaries to reflect population and voting patterns rather than land area.

deck.gl Python GeoPandas Anamorphic Mapping
Planning

Windsurfing App

App which is providing Windsurfers with actual spot informations - to make it easier to decide to go or not to go and which gear has to be chosen

App Development Frontend Design Community Building Marketing
Planned

Llama2Knowledge

A fully local, AI-driven knowledge base. Feeds a personal library of markdowns and PDFs into a vector store, builds a searchable knowledge graph via Ollama inference — no cloud, no external dependencies, no data leaving the machine.

Local AI Knowledge Graph RAG Ollama Vector Search Self-Hosted
Planned

WarSAM

Using Segment Anything Model to detect crater and grenade scars on Sentinel imagery — tracking frontline evolution and combat activity in Ukraine through open-source satellite data.

OSINT Remote Sensing Sentinel SAM3

Recent Writing

Thoughts on building, failing, and figuring things out

Chasing the Storm
April 2026 · Personal

Chasing the Storm

Why I started (and stopped) a YouTube ambient sound channel — and what the experiment was actually worth.

Read more →
Coming Soon · Geospatial ML

Common Pitfalls in Geospatial ML Projects

Real-world mistakes I've seen (and made) when applying machine learning to satellite and spatial data — and how to avoid them.

Coming soon
Coming Soon · Career

From Academic to Industry: What I Wish I'd Known

Lessons learned making the transition from PhD research to tech leadership — the things nobody tells you before you make the jump.

Coming soon

Learning Journey

How my skills evolved through different phases of my career

Spatial Intelligence Lead
Architecting scalable AI systems and leading technical teams. Balancing B2B/B2G enterprise requirements with high-performance ML engineering, from custom transformer training to distributed inference pipelines.

Leadership & Strategy

System Architecture
Cross-functional Team Lead
B2B & B2G Consulting

AI Engineering

Computer Vision (Mask2Former)
PyTorch (DDP, AMP)
GenAI & Agentic Systems

Infrastructure & MLOps

Ray Distributed Inference
Docker & CI/CD
PostgreSQL / PostGIS

Key Projects

Automated Inference Pipeline
Led design of a distributed company-wide inference pipeline (TTA, dynamic queue loading) serving multiple segmentation tasks with scalable Ray architecture.
Ray GPU Optimization System Architecture
Mask2Former: Photovoltaics Potential
Developed production instance segmentation prototype for rooftops handling multi-channel inputs (RGBI) with custom Tanimoto/Boundary-aware loss functions.
PyTorch HuggingFace Custom Loss Design
Company-Wide Operations Scaling
Built and implemented comprehensive PM systems (Monday.com) aligning agile development cycles with rigorous B2B client deliverables.
Agile Processes Team Leadership
On-Premise AI Lead Evaluation
Engineered a localized agentic AI prototype utilizing Ollama for automated semantic enrichment of B2B data securely.
Self-Hosting AI LLM Orchestration
Industry Transition
Engineered production-ready geospatial ML pipelines. Focused on translating nationwide spatial datasets (OSM, Sentinel-2) into viable data products under real-world computational constraints.

Machine Learning

Deep Learning (Segmentation)
Time Series & Predictive
Spatial Python (GeoPandas)

Production Systems

Docker Environments
MLflow Tracking
Pure SQL Spatial Pipelines

Professional Skills

Product Development
Client Communication
Technical Consultation

Key Projects

Renewable Energy Potential (PV & Wind)
Developed nationwide spatial derivation models calculating Global Solar Irradiation and Weibull wind distributions integrating ALKIS cadastral data.
PostGIS Spatial Math GeoPandas
Sentinel-2 Land Cover & COG Mosaics
Deployed large-scale classification models and processed massive Cloud Optimized GeoTIFF (COG) VRT mosaics with Ray parallelization.
PyTorch Rasterio GDAL
High-Resolution Forest Segmentation
Built custom PyTorch models to classify wood types from aerial imagery, advancing from UNet baselines to transformer architectures.
Computer Vision MLflow
PhD in Geoinformatics
Constructed big-data harvesting pipelines and applied NLP/Deep Learning algorithms to extract structured spatial intelligence from social media streams. Strongly focused on connecting empirical citizen science with urban planning infrastructures.

Data & AI Algorithms

NLP & Text Mining (TF-IDF)
Predictive Modeling (CNN, SVM)
API Data Harvesting

Spatial Infrastructure

GeoServer / SLD
PostGIS Database Design
QGIS & Desktop GIS

Research & Citizen Science

Empirical Research
Survey Design
Teaching & Lecturing

Key Projects

Spatial Event Analytics (Hanse Sail & Protests)
Evaluated spatial-temporal dynamics and sentiment during major urban events (Hanse Sail, demonstrations, DFB-Pokal) via real-time social media harvesting to derive actionable insights for urban security.
Spatial-Temporal Analysis Sentiment Analysis Urban Intel
Citizen Science: KOGGE Urban Water Surveys
Conducted comprehensive empirical online surveys (willingness-to-pay analysis) to connect citizen participation directly with municipal hydrological infrastructure planning.
Citizen Science Survey Design Empirical Research
Dissertation: Spatial NLP Pipeline
Engineered a full-stack system extracting spatial entities from raw Twitter API streams utilizing Word2Vec, CNNs, SVMs, and multi-layer gazetteer-matching for digital civic participation.
NLP Deep Learning Data Engineering
KOGGE: Hydrological Data Infrastructure
Managed database architecture creating fully INSPIRE-compliant spatial systems (GeoServer/GeoNetwork) for regional catchment analysis.
SDI PostGIS System Architecture
Employee for Data and Services
Applied academic foundations to legacy commercial environments. Established foundational understandings of municipal data requirements and rigorous quality control protocols.

Foundational Tools

MapInfo
Visual Basic
Data Processing

Professional Core

Quality Control
Client Communication
Project Planning

Key Projects

Automated Analog Data Processing
Developed scripting logic to digitize, correct, and unify analog municipal sewer and freshwater system records.
Algorithm Design Municipal Data GIS
Master of Science in Geoinformatics
Established deep domain expertise in spatial analysis, environmental monitoring, and theoretical cartography.

Domain Foundation

Field Measurements
Hydrological Modeling
Spatial Statistics

Early Technical Base

ArcGIS
Remote Sensing Basics
Python Fundamentals

Key Projects

Tangra Yumco Hydrological Modeling
Executed system analysis and complex time series modeling of a high-altitude catchment area for thesis defense.
Time Series Analysis Climate Data ArcGIS
Thuringian Forest Kiln Analysis
Conducted comprehensive field extraction and laboratory analysis assessing carbon retention properties.
Field Work Scientific Writing
River Quality Campaign
Collected and statistically analyzed velocity and chemical distribution across local river networks.
Data Collection Statistical Analysis

Skills Overview

Comprehensive skill set built across research, engineering, and leadership

AI & Machine Learning

Computer Vision (Mask2Former, SegFormer)
PyTorch (DDP, AMP, Custom Loss Design)
NLP & Generative AI (LLMs, RAG, Word2Vec)
Scalable Inference (Ray, TTA, GPU Opt.)
Predictive Modeling (CNNs, LSTMs, SVM)

Spatial Intelligence & GIS

Spatial Python (GeoPandas, Rasterio, GDAL)
Remote Sensing (Sentinel-2, SAR, COG, ESA SNAP/NEST)
Spatial Data Infrastructure (GeoServer, SLD, INSPIRE)
PostGIS & Pure SQL Spatial Pipelines
WebGIS & Spatial Algorithms (deck.gl, Anamorphosis)

Software Engineering & MLOps

Core Languages (Python, JavaScript, SQL, Java)
Cloud & Data Eng. (AWS S3, PostgreSQL, FastAPI)
MLOps (Ray Distributed, MLflow)
Docker (CUDA environments, Compose)
GitLab CI/CD & Pipeline Automation

Research & Leadership

System Architecture & Technical Strategy
Project Delivery (50+ Enterprise & Public Sector Projects)
B2B & B2G Consulting (PwC, E.ON, Ministries, Municipalities)
Empirical Research & Survey Design (Citizen Science)
Cross-functional Team Leadership

Media & Technical Communication

3D Simulation & Modeling (Blender)
Video Post-Processing & Color Grading (DaVinci Resolve)
Content Creation & Distribution (YouTube)
Technical Writing & Documentation
Data Storytelling & Visualization