Skip to main content

Senior Data & AI Platform Architect

Thorsten Born

Germany-based senior architect with hands-on technical leadership across enterprise data platforms, dimensional and Data Vault modeling, lakehouse architecture, and production AI.

Contact About a Role View Case Studies

Target roles

Data Architect · Data Platform Architect · AI Platform Engineer · AI Architect · Technical Lead, Data & AI Platforms

Location

Germany, DACH, Europe, compatible global remote

Languages

German native, English fluent

Profile

I design data platforms and the AI systems that run on them. My work covers dimensional and Data Vault modeling, lakehouse and warehouse design, streaming and batch pipelines, retrieval with source attribution, agentic workflows with explicit human review points, and multi-tenant infrastructure on Kubernetes. I also created and maintain Caudra, an open source terminal coding agent written in Rust. I do my best work in senior hands-on roles where one person connects data modeling, architecture, implementation, technical direction, mentoring, evaluation criteria, and the handover to operations.

Professional Experience

AI Engineer & Architecture Consultant

Independent

Aachen, Germany | 2025 - Present

  • Deliver architecture advisory for enterprise data platform, AI/ML, and IoT initiatives, translating established workflows into target architectures and rollout plans
  • Advise on lakehouse and warehouse target architectures, including modeling standards, transformation layers, and migration sequencing away from legacy warehouses
  • Design and facilitate enterprise AI workshops that turn LLM and retrieval use cases into evaluation criteria, guardrail requirements, and measurable adoption plans
  • Cover agentic workflow patterns, Node-RED, Kubernetes, observability, human review boundaries, and operational handover

Senior Data Architect

EdgeIQ

Remote (US) | 2023 - 2025

  • Owned architecture and hands-on delivery for a multi-tenant enterprise IoT workflow platform, reducing customer onboarding time by 70% through reusable workflow and integration patterns across 100+ custom Node-RED nodes
  • Owned the tenant-facing data model across device, event, and time-series domains, using dimensional modeling and a dbt transformation layer to keep analytics contracts stable while the underlying IoT schema evolved
  • Designed Generic Timeseries API (Go/TimescaleDB), replacing 5+ specific endpoints and cutting dashboard delivery from weeks to hours
  • Automated multi-tenant Kubernetes provisioning, eliminating an estimated 40+ hours/month of manual DevOps with 1-second reconciliation
  • Standardized workflow runtime with 50+ documented device-lifecycle subflows and per-tenant isolation, with an architecture target of 1,000+ workflow instances
  • Built 10+ React dashboard components with custom hooks for real-time IoT fleet visualization
Related case study EdgeIQ

Data Scientist & Data Engineer

HDI (Talanx Group)

Cologne, Germany | 2020 - 2022

  • Led development of a German-language NLP pipeline automating >95% of bank-data change email requests and saving an estimated 200+ analyst hours/month
  • Integrated Planet AI handwriting recognition and validation workflows processing 15,000+ forms/month, reducing manual transcription in insurance document operations
  • Modeled analytics domains against HDI's enterprise Data Warehouse using Kimball dimensional and Data Vault 2.0 patterns, giving downstream data science teams a consistent consumption layer
  • Built Python/PySpark ETL bridges from HDI's enterprise Data Warehouse and legacy insurance systems into reusable analytics workflows
  • Presented predictive analytics strategy to Life Insurance Board Member (Vorstand)
  • Established ML engineering standards adopted by 3 data science teams / 12+ engineers

Selected Projects & Ventures

Creator & Maintainer (Open Source)

Caudra

Aachen, Germany | 2026 - Present

  • Created and maintain Caudra, an open source terminal coding agent written in Rust and forked from Maki
  • Built the features I wanted from plugins into the agent itself. It ships steerable subagents, built-in token savings, a workbench, reversible history, Plan mode, and /goal in one native binary
  • Runs on macOS, Linux, and Windows and works with 17 built-in model providers and with local models
  • Use it for my own work every day and push more than 20 billion tokens a month through it (my own usage, not a benchmark)
  • Released as a free, open source public preview
Related case study caudra.ai

Port & Maintainer (Hobby Project)

NInfer-4090

Aachen, Germany | 2026 - Present (Hobby)

  • Ported NInfer-3090, the fork by Don-Chad of the NInfer engine by Neroued, to the RTX 4090 to run Qwen3.8-27B with the full 262,144-token context on one 24 GB card
  • Added per-request LoRA adapters and a decisions API on the chat model's weights, which Caudra uses as its local decision engine
  • Measured 3,600 tok/s prefill on a 9,411-token prompt and 148.6 tok/s code decode on one RTX 4090
Related case study ninfer-4090 on GitHub

Founder & Lead Developer (Side Project)

Zeitgaist

Aachen, Germany | 2023 - Present

  • Started immediately after ChatGPT's first public launch as a Perplexity-like real-time RAG answer engine while mainstream assistants were still knowledge-cutoff-limited
  • Built dense retrieval with Sentence Transformers and cross-encoder reranking across 6 social platforms
  • Implemented cross-lingual search supporting 20+ languages with automatic query translation
  • Designed real-time WebSocket streaming with MessagePack serialization for AI chat responses
  • Deployed scalable infrastructure on Docker Swarm with Redis caching and OpenTelemetry tracing
Related case study zeitgaist.ai

Co-founder & Lead Developer (Startup)

Foretale

Aachen, Germany | 2021 - 2024

  • Co-founded no-code trading platform with visual workflow automation for cryptocurrency strategies
  • Built distributed ML backend (FastAPI/Ray/PyTorch) for real-time sentiment analysis and OCR
  • Created custom Node-RED fork with 25+ trading nodes and Svelte-based UI components
  • Designed Kafka-based sensor network processing Twitter/whale/sentiment data 24/7
  • Modeled the analytical store behind the strategy backtester, separating raw event capture from a dimensional layer so signal research and live execution read consistent definitions
Related case study foretale.io

Earlier Experience

Consultant & Developer (part-time)

TurnDigital GbR -- IT Consulting

Aachen, Germany | 2019 - 2022

  • Delivered ML forecasting and AWS infrastructure modernization for SMB clients, contributing to reported reductions of 15% in inventory costs and 30% in cloud costs

Software Engineering Intern

Silexica GmbH -- later acquired by Xilinx, now AMD

Cologne, Germany | 2017 - 2018

  • Built Yocto/BitBake integration for SLX Tools across 6 target architectures, enabling analysis of embedded Linux applications for automatic parallelization
  • Analyzed an AUTOSAR Adaptive autonomous-driving workload and identified an analyzer-estimated 8x theoretical speedup ceiling on an 8-core platform
  • Supported 6 processor architectures (ARM, MIPS, PowerPC, x86) for cross-platform analysis
Xilinx acquisition announcement

Working Student -- R&D

Lumileds Germany GmbH (formerly Philips)

Aachen, Germany | 2014 - 2017

  • Conducted hardware measurements and stress testing for automotive LED products and developed firmware optimizations

Engineering Intern

Halfkann + Kirchner

Germany | Summer 2007

  • Designed electronic control system for patented fire simulation device (DE102008011567B4)
  • Built custom PCB and programmed ATmega microcontroller in C/Assembler
  • Created Windows GUI enabling engineers to draw custom fire progression curves
  • Developed DLL driver bridging hardware and software
Related case study Patent

Education

Master of Science in Electrical Engineering, IT, and Computer Engineering

RWTH Aachen University

Aachen, Germany | 2018 - 2020

  • Focus on Machine Learning, Embedded Systems, and High Performance Computing
  • Developed a hybrid EA+RL framework combining behavior trees for instinctive behavior with Q-learning, PPO, and DQN for learned behavior
  • Thesis: Evolving Behavior Trees for Reinforcement Learning on Medical Databases (1.0, best possible German grade)
  • Related paper: EVO-RL: Evolutionary-Driven Reinforcement Learning ( arXiv:2007.04725) View paper

Bachelor of Science in Electrical Engineering, IT, and Computer Engineering

RWTH Aachen University

Aachen, Germany | 2013 - 2018

  • Focus on Computer Engineering, High Performance Computing, and Digital Systems
  • Implemented an SHM-LXC pscom plugin for Parastation MPI, enabling shared-memory communication between containerized HPC nodes
  • Thesis: Implementation of a Light-weight Software Layer for Inter-process Communication (1.3, second-best possible German grade)