OpenAI Experts in Munich

• This project involved extraction of intelligence data to support report writing for a company that provides geopolitical, global, commercial intelligence. The data have been gathered from a number of resources (interview transcripts, online data, internal documents), and then a knowledge base has been build from it. This was the basis of a complex RAG system, that was evaluated against a golden dataset. Agents have been used to find out the contradicting intelligence, the statements supporting each other, and to store back the generated knowledge.

Used: Python, RAG, LangGraph, LangChain, deepeval, MCP

• Recurring problem in my own AI-assisted projects: requirements analysis, use cases, and architecture decisions can be created quickly with AI support, but remain difficult to follow and scattered across Markdown files – knowledge is lost as soon as it is no longer in the context window. arknet turns requirements engineering and architecture knowledge into structured, verifiable data instead of plain text: requirements, use cases, and architecture decisions form a consistently linked knowledge graph, traceable from requirement to architecture decision – queryable by both people and AI agents. Technically based on RDF/OWL and a custom MCP server.

• Result: Working MCP daemon, Docker image published automatically to GHCR, nine hexagonal modules, eleven ADRs (including an Open-Core licensing model). Requirements engineering and Ubiquitous Language hexagons are active. Public as a Community Edition under Apache-2.0 since 07/2026 (github.com/kogn-io/arknet), together with the Claude Code plugin and GHCR image; Open-Core model.

• Label: Java, Maven, RDF, RDF4J, OWL, SPARQL, Model Context Protocol, Spring AI, Docker, GitHub, Git, Claude Code, Obsidian, DDD, Hexagonal Architecture, ArchUnit, JUnit, AssertJ, Interface Development, Software Architecture, Continuous Integration, Knowledge Management

• Modernization of a large-scale Java enterprise application*

The project is modernizing a complex enterprise application that has grown over many years. The existing Spring-based legacy system runs on Java 8, OSGi, and Eclipse RCP and is being gradually migrated to a modern, maintainable architecture with Java 25 and Quarkus.

Marcus works on analysis, architecture, refactoring, and implementation. One focus is on untangling historically grown structures and dependencies and on building a clean, sustainable Java and Quarkus technology stack.

Tools & technologies: Java 8, Java 25, Quarkus, Hibernate ORM with Panache, EclipseLink, OSGi, Eclipse RCP, Maven, JUnit, Mockito, REST, JSON, Git, Eclipse IDE, IntelliJ IDEA Ultimate, Jira, Confluence

• Overall responsibility for the project lifecycle from scope definition to completion
• Close collaboration with platform teams, IT leaders, and external service providers
• Application of SAFe principles and structured sprint work
• Creation of a migration roadmap with clear milestones
• Monitoring of the lifecycle: onboarding, repository migration, replication of permissions, and system tests
• Visualization of the architecture with PlantUML and Gliffy as well as documentation in Confluence
• Regular status reports and running knowledge transfer sessions

• Ongoing mandate: advisory and business case development for digital services, from the profitability model through to KPI definition for the rollout.

Loom and Bloom

Python · TypeScript · n8n · Claude Code · Whisper · Gemini · Supabase · Notion · HubSpot · Digital Ocean

• Built an end-to-end content pipeline: one Loom video → marketing images, bilingual LinkedIn posts, newsletter and Help Center updates.
• n8n webhook → SSH → Claude Code session on a Digital Ocean VPS; three MCP servers (video, Notion, Supabase).
• Whisper word-level transcription, ffmpeg screenshots, Gemini UI annotation, PIL device mockups.
• Next.js upload UI plus a bilingual newsletter composer with HubSpot push.

AI Assistance and Target Vision for Partially Autonomous Energy Portfolio Management

Building an AI control layer directly on the up-to-date daily live portfolio of a large energy provider — not as an isolated pilot, but as an operational extension of the existing DB1 and portfolio management. The goal is the gradual development from assistance through monitoring/alerting to analysis agents with Human-in-the-Loop approvals, supplemented by a role-specific System of Engagement alongside the BI System of Record. At the same time, the business case, target vision and management pitch compared with static monthly reporting are being developed.

• AI control layer: Design and build on the existing live portfolio data product (several million contracts) — development stages assistance → monitoring/alerting → agents with Human-in-the-Loop approvals.
• LLM-supported data analysis: Semantic queries, SQL/tool integration and additional RAG components based on portfolio, plan-versus-actual and data quality data (Azure OpenAI, Snowflake), with drill-downs to individual contract level.
• Analysis agents: Multi-stage agents for variance and driver analyses of churn, price adjustments, volumes and procurement costs.
• Views concept: Role-specific interfaces for business units, management and C-level as a System of Engagement alongside the BI System of Record.
• Business case & pitch: Target vision and cost-benefit argumentation compared with static monthly reporting.
• LLM setup (privacy & security): Coordination with IT Security and Data Protection — EU region, data separation and approval processes.
• Agent architecture: Multi-stage agent pipelines (analysis → validation → summary) with documented data sources, tool calls, review steps and source references for each statement.
• Data foundation: Built on the up-to-date daily DB1 data product (Snowflake, dbt) — portfolio, plan-versus-actual and data quality metrics as the common basis for all AI analyses.
• Guardrails & evaluation: Evaluation and approval processes for LLM responses relating to management-relevant statements — test sets, metrics and human review.
• Prototyping & validation: Iterative validation of agent responses with the business unit — test question catalogue, feedback loops and response quality for each release.
• Roadmap & development stages: Detailed stages from assistance → monitoring/alerting → partially autonomous management, including transition criteria and governance for each stage.
• Integration: Integration into the existing BI and data landscape — BI remains the System of Record, while the AI layer provides interactive drill-down paths as the System of Engagement.
• Enablement: Enablement of business users — prompting guides, training and an operating model for ongoing use.
• Management: Coordination of business units, Data Engineering, IT Security and Data Protection.
• Change Management: Communication and expectation management with business units and management throughout the development stages.

Results:

• Built on an existing up-to-date daily data product with several million contracts
• Established an LLM setup coordinated with Data Protection and IT Security in the EU region, including data separation and approvals
• Defined three development stages through to partially autonomous management
• Designed role-specific views for business units, management and C-level
• Developed the business case and management pitch for the development stages
• Established an iterative response-quality validation process with the business unit
• Designed the operating model for assistance operations and initiated validation

Stack: Azure OpenAI, Azure AI Foundry, Snowflake, dbt, React, TypeScript, Entra ID, RAG, Agentic AI, analysis agents, Human-in-the-Loop, Prompt Engineering, LLM Evaluation, LLMOps, GDPR / EU region, Azure DevOps, Python, SQL, Change Management

AI-driven automation of purchasing on a printing company's website, including selecting delivery times, order options, ordering, payment, and uploading print data from the Lumiz Cloud.

• Research and develop models for chatbots, NLP and LLMs (e.g. Llama, Qwen, OpenAI)
• Enhance chatbots with RAG, in-context learning
• Supervised fine-tuning (PEFT, LoRA), Huggingface or Unsloth
• Advanced training methods: Test-time training, (transductive) active learning, reinforcement learning
• High-throughput serving with vLLM
• Apply embedding models (e.g. SentenceTransformers), similarity/vector search or vector DB or ranking (e.g. LlamaIndex, Faiss, LangChain)
• Generate and filter synthetic data, clustering
• Detect hallucinations
• Evaluate chatbot models (Rouge, BLEU, F1-Score, Recall, Precision)
• Visualization of experiments (matplotlib)

• Completed an intensive AI Engineering Program focused on LLM evaluation, retrieval, agent orchestration, and multimodal workflows.
• Designed retrieval pipelines with document ingestion, semantic search, and citation-aware outputs using LangChain and ChromaDB.
• Built LangGraph-based agent workflows with state handling, clarification loops, and human-in-the-loop approval steps.
• Applied prompt engineering and evaluation patterns, including scoring logic and guardrails, to improve output quality and reliability.
• Worked extensively with Python, FastAPI, OpenAI APIs, LangChain, LangGraph, and ChromaDB in end-to-end implementations.

• Integrated LangChain middleware between AEM and SAP PIM system
• Developed a FastAPI interface for system communication
• Implemented vector embeddings for semantic product search
• Evaluated LLM models (Vertex AI/Gemini, LM Studio, Hugging Face, OpenAI) for product analysis
• Developed an AEM component to display product recommendations and integrated the recommendation API into the AEM authoring process
• Designed and implemented Pinecone vector database for product embeddings
• Optimized response times and caching strategies
• Evaluated Vertex AI Studio for LLM testing and prompt workflows
• Implemented secure API routing and access control for AI components via FastAPI and gateway validation

• Concept creation and implementing AI Agents in AWS Cloud
• Continuously alignment with stakeholders
• Collaborate with DevOps
• Technologies: Git, CI/CD (GitHub Actions), Python/ML, Streamlit, Deno/typescript, AWS SAM, AWS Bedrock, AWS Lambda, AWS Dynamo DB, AWS S3, AWS Event Bridge etc.

• Clarification of project goals, scope, and functional target vision for building the AI-based knowledge base.
• Deriving the initial architecture and implementation strategy for the Zasterbot chatbot, including defining the MVP and expansion phases.
• Developing a functional target vision for building a structured knowledge base and integrating a future chatbot.
• Deriving and prioritizing use cases for information retrieval and provision by the chatbot.
• Modeling data structures and flows for effectively organizing the knowledge base on the Base44 platform.
• Designing and implementing data models for storing and linking relevant information.
• Developing processes for extracting, analyzing, and preparing raw data for the knowledge base.
• Ensuring data consistency and quality as the foundation for the future chatbot.
• Planning the integration of large language models (LLMs) and retrieval-augmented generation (RAG) for precise and context-aware responses.
• Implementing features for analyzing and visualizing data from the knowledge base.
• Using the Base44 platform with JSON-schema-based entities and a flexible permission model.
• Implementing Deno functions for backend logic, event processing, and external API integration.
• Integrating OpenAI services for initial data analysis.

• Co-founded an AI venture focused on building production-grade LLM applications and agentic systems
• Designed and implemented multi-agent AI workflows for financial and trading applications
• Developed LLM-powered copilot architectures for portfolio analysis, trade management, and personalized user coaching
• Built on-device and edge-deployed inference applications, optimizing models for low latency, privacy, and resource-constrained environments
• Led system architecture decisions across model selection, orchestration, state management, and deployment