Building resilient agentic development processes through context engineering, clear specs, modular skills and workflows, specialized subagents, and automated validation loops.
Evidenced by
01 / Profile
Agent-first with engineering foundation.
Autonomous execution requires engineering control over architecture, code quality, and technical risks. This control draws on a mathematical foundation and eight years of experience in software and systems engineering for safety-critical applications. This combination enables reliable, practical AI solutions.
Core Competencies
Developing software for data-driven, embedded, and full-stack systems, designing architectures, data flows, and interfaces, and safeguarding quality through type safety, automated tests, and CI/CD.
Evidenced by
Building local LLM infrastructure, offline RAG, and secure interfaces to protect sensitive enterprise data and operate AI workloads with data sovereignty inside the company network.
Evidenced by
Evaluating the capabilities and limitations of AI models, structuring tasks through precise context and prompt design, and systematically assessing results against defined criteria.
Evidenced by
Evaluating development environments and toolchains, embedding best practices for agentic engineering across the team, and enabling colleagues to use AI-assisted development tools productively.
Evidenced by
02 / Systems
Systems instead of buzzwords.
Developed and orchestrated with Claude Code, Codex, GitHub Copilot, and Antigravity.
Focus Projects
Company Brain
Tool-agnostic agent harness for engineering teams to structure global standards, private workspaces, and repositories with modular skills, workflows, and MCP integrations.
Tech Stack & Architecture
Agent Harness · Context Engineering · Skills, Workflows & Hooks · MCP · Tool-Agnostic Design
Local RAG System
Containerized offline RAG system for company-wide document chat with local open-source models on dedicated GPU infrastructure under complete data sovereignty.
Tech Stack & Architecture
RAG · Local LLMs · HuggingFace · GPU Infrastructure · vLLM · Docker · Kubernetes · Helm · Data Sovereignty
GiftFlow
Agentically engineered full-stack web application for occasion and gift cards with Supabase RLS, Stripe and Tremendous integration, 3D canvas, and Playwright E2E tests.
Tech Stack & Architecture
Next.js · TypeScript · Supabase RLS · Stripe API · Tremendous API · Three.js · Playwright
Development Systems
Personal Agent Workspace
Personal agent harness and knowledge architecture: Manages project knowledge, specs, and session handoffs with automated learning capture.
Tech Stack & Architecture
Agent Harness · Context Engineering · Knowledge Architecture · Obsidian · Feedback Loops
Home Assistant
Automated control of networked smart home devices based on internal and external sensor data (occupancy, indoor climate, weather data) within an agent-maintained IoT system.
Tech Stack & Architecture
YAML · Jinja2 · Zigbee2MQTT · ESP32 · Automation Engine
Applied AI
Voice AI Agent
Event-driven real-time voice agent for autonomous candidate interviews with webhook-based post-processing, structured transcript analysis, and automated Slack reporting.
Tech Stack & Architecture
Voice Agents · Vapi · n8n · Webhooks · Structured Outputs
ML-Powered CRM
Full-stack lead management platform featuring a FastAPI REST API, React dashboard, Docker setup, and ML models (PyTorch, TensorFlow) for lead status prediction.
Tech Stack & Architecture
Python · FastAPI · React · Machine Learning · PyTorch · Docker
AI Enablement & Automation
GitHub Copilot Enablement
Introduced GitHub Copilot across the development team, established best practices for AI-assisted workflows, and trained colleagues.
Tech Stack & Architecture
GitHub Copilot · Enablement · AI Tooling
Process Automation
Systematically automated recurring engineering, simulation, and evaluation workflows via Python-based tooling pipelines and CI/CD integrations.
Tech Stack & Architecture
Python · CI/CD Automation · Developer Tooling · Data Pipelines
03 / Operating Model
Steer context and agents. Own the results.
- Spec before code.Objectives and acceptance criteria are fixed upfront.
- Autonomy is bounded.Permissions and stop conditions are clearly defined.
- Evidence decides.Tests and independent review substantiate quality.
Set up the context system
ApproachSet up task-focused workspaces and reduce memory files to the universal minimum. Structure project rules, conventions, and toolchains hierarchically, while loading domain-specific workflows on demand via skills and MCP integrations.
Quality gateThe harness stays lean, token context remains focused, and every session starts with precisely the required capabilities.
Define goals and boundaries
ApproachUse plan mode to research and analyze complex tasks or architectural decisions in isolation before modifying code. Clarify objectives, permissible autonomy, gated actions, and stop criteria to prevent context pollution from trial and error.
Quality gateThe approach is understood before touching code, autonomy boundaries are clear, and critical actions require explicit confirmation.
Derive the spec and validation plan
ApproachStrictly separate planning from building by creating a versioned feature spec prior to implementation. Based on documented project memory, objectives, technical constraints, and measurable acceptance criteria including the test plan are established up front.
Quality gatePlanning and building are decoupled, the spec is unambiguous, and the verification method for every acceptance criterion is established.
Orchestrate implementation
ApproachBreak down the spec into small, verifiable work packages and delegate implementation to specialized subagents. Acting as orchestrator, the engineer governs interfaces, context, and dependencies while agents build and integrate incrementally.
Quality gateWork packages are clearly bounded, dependencies are cleanly decoupled, and every increment is immediately runnable and integrable.
Validate independently
ApproachVerify the implementation in a fresh context against the spec acceptance criteria to eliminate confirmation bias from the authoring agent. Automated test suites, type checking, and targeted edge-case reviews provide reproducible evidence for final sign-off.
Quality gateEvery acceptance criterion is backed by concrete evidence, side effects are ruled out, and release decisions rely on verified quality.
Embed knowledge
ApproachRecord architectural decisions and concepts permanently in project memory. Key learnings flow directly into agent rules or automated tests, while temporary specifications are archived to prevent context drift and technical debt.
Quality gateCode and documentation remain in sync, the system systematically learns from mistakes, and the harness stays lean for future sessions.
04 / Experience
Engineering for production systems.
AI / ML Software Engineer Video / Image
M4Com SYSTEM GmbH
Python software development for video and imaging systems focusing on local AI infrastructure, RAG systems, and coding agents.
- Architected local AI infrastructure and hardware sizing (NVIDIA H200) and built coding-agent workflows
- Advanced data-sovereign AI workloads from evaluation to production operation
R&D Engineer SW Development
ZF Group
Modular software for engine control units and systematic automation of development and validation workflows.
- Systematically automated development and simulation workflows with Python
- Established AI development in the team and shared knowledge
- Implemented CI/CD pipelines in Azure DevOps for automated builds and tests
R&D Engineer SW & Function Development
FERCHAU · Deployment at ZF Group
Software and function development for smart transmission diagnostics.
- Developed and data-calibrated diagnostic functions for passenger car automatic transmissions
- Validated through SIL, HIL, and in-vehicle testing
05 / Foundation
Analytical depth, proven stack.
Track record in data-driven, embedded, and safety-relevant systems: architecture, full-stack, data processing, APIs, testing, and CI/CD.
- Python
- TypeScript
- FastAPI
- React
- PyTorch
- vLLM
- SQL
- Supabase
- MATLAB & Simulink
- Azure
- Docker
- Linux
- CI/CD
- Git
- C
Education & Professional Development
- M.Sc. TechnomathematicsTechnical University of Berlin · 04/2015-04/2018
- B.Sc. Business MathematicsUlm University · 10/2011-04/2015
- AI ManagerEverlast AI · Certificate ↗
- Claude Code MasterclassEverlast AI · Certificate ↗
- Agentic Coding MasterclassEverlast AI · Certificate ↗