Multi-Agent Systems Experts in Germany

• Claude Code subagents, MCP, Pydantic V2, pytest, bandit

• Designed and shipped 3 specialized agents that hand work down a line: a research agent writes a cited implementation spec, a coding agent builds the modular code and its tests, a review agent ranks findings by severity and applies the fixes. Each handoff is a structured document, so no stage depends on another agent's context window.

• Connected the research agent to an academic-research MCP server (Semantic Scholar, ArXiv, Hugging Face Hub, citation snowballing) so every reference traces to a tool result rather than the model. Gated commits behind ruff, mypy, pytest and bandit, required human sign-off before installs and commits, and persisted session state on disk so long runs survive a context reset.

Project: Defect Generation on Test-Bench Images of Metal Surfaces Environment: Automated Visual Inspection (AVI), Metallurgy & Manufacturing

• Objective & Implementation: Designed, architected, and trained Generative Adversarial Networks (Pix2PixHD / SPADE) for image-to-image transformation. Targeted generation of synthetic material defects (e.g., cracks, inclusions, scale) on rough metal surfaces under real test-bench lighting conditions for privacy-compliant and efficient dataset expansion (data augmentation).
• Technical Design: Implemented robust Generative AI and computer vision pipelines in Python and PyTorch. Used semantic segmentation approaches for mask-controlled defect synthesis and subsequent evaluation with EfficientDet object detection models.
• Business Impact: Massive dataset upscaling (10x) without time-consuming and costly physical test-bench runs, while significantly improving the detection performance of automated inspection systems.

Technologies & Skills Used: Python | PyTorch | SPADE | Pix2PixHD | EfficientDet | Machine Learning | Semantic Segmentation | Computer Vision

Stabilization of an Azure/.NET landscape in live operation.

• Architecture, DevOps, and operational readiness; technical decisions under time pressure
• Azure DevOps, monitoring, ETL/ELT, cloud security, FinOps, and data-mesh-related topics

Technologies: Azure DevOps, .NET, CI/CD, monitoring, FinOps

Multi-Agent Document-Generation Platform

Problem:

Turn a single prompt into a well-researched, structured and cited document, while keeping the user in control: every stage can be reviewed, corrected or revisited before the system moves on.

Architecture:

A multi-agent system orchestrated by a central controller. More than 20 specialised pydantic-ai subagents (understanding, structuring, planning, research, writing, assembly, cross-reference check, glossary, citation, reference table and more) pass results forward through a shared plan persisted in Postgres (asyncpg); a runner layer dispatches each task by name to its agent, and mechanical steps such as formatting, table of contents and list of figures are deterministic code instead of LLM calls. Provider (Anthropic or OpenAI) is a configuration setting, and responses stream over Server-Sent Events.

Hosting & Infrastructure:

Stateless container run as a Scaleway Serverless Container on a private network with managed PostgreSQL, object storage for finished documents and assets, and centralized log shipping (Loki / Scaleway Cockpit), provisioned with OpenTofu modules and deployed through GitLab CI; usage is metered per API key.

My Contribution & Key Decisions:

• Central decision controller: a satisfaction classifier turns the user's free-text feedback into a verdict (keep working, advance, cancel or revise an earlier task), and a middleware applies it to the plan. Plan progression supports jumping back to earlier tasks and re-running the stages after them.
• Research and writing fan-out: planning produces a plan from which a research-design agent generates individual research tasks and a writing-design agent generates individual writing tasks, each linked to the research it needs. Every task runs in its own dedicated agent with its own conversation history.
• Self-review and human-in-the-loop: after each step a critical-thinking agent reviews the output; the user can accept, refine, or ask questions about a critique before the plan advances.
• MCP integrations: two remote MCP servers, web search and internal document search, connected through pydantic-ai with retries and long-running tool-call timeouts. Classifier agents decide per task whether web search is needed, and a small model condenses raw search results into structured findings before they reach the writing agents.

Usage Metering:

Clients authenticate with an API key validated against a key-management service; consumption is charged in credits calculated from token usage per model plus a markup, and exhausted quota is answered with HTTP 429.

• 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

• Built an MCP (Model Context Protocol) layer for Mwanachama's agency service, turning domain manager methods into callable AI-agent tools. This included a composite tool that builds a full organization design (org chart, goals, workflows, RACI matrix) from one specification.
• Built the chat-driven agency builder (Wakala Studio and API), where an organization describes its structure in natural language and an AI agent uses those tools to construct and modify the live design.
• Added an insights service so an organization can review AI-agent interactions and completed work. Insights from that review feed back into solution design, gated by architect and user sign-off.
• Alongside this, designed and built the platform itself: ~20 Go microservices on PostgreSQL, Flutter and React clients, deployed on Kubernetes.
• AI agents scan the platform autonomously for security gaps and run scripted tests, covering API (Postman-style) and UI testing. The rest of the work stays supervised. No rogue agents, promise.
• Building an AI-native enterprise platform that structures organizational data, workflows and responsibilities for both people and AI agents.
• Models accountability through RACI, allowing human and AI participants to take different operational roles, including Responsible, Accountable, Consulted and Informed. AI agents can also take controlled approval roles where the organization assigns that responsibility.
• Extended the organizational and operational-risk themes from my Executive MBA thesis into software, making roles, processes and controls explicit and machine-readable.
• Structures enterprise context so LLM interactions use more relevant information instead of repeatedly processing large amounts of unstructured content.

• Task Engineering: Evaluating Coding Agents.
• Asset Creation: Building realistic task environments using datasets and files in German. Crucially, these assets must remain in the target language to genuinely measure multilingual handling.
• Prompting & Translation: finding failure points where AI does not work, in German.
• Implementation & Verification: Supporting the development of robust solutions (reference implementations) and write highly reliable, deterministic verifier scripts (using rubric-based judging only when strictly necessary).
• Calibration & Execution: Analyze execution logs and calibrate task difficulty (Easy to Very Hard) using standard Terminal-Bench run configurations against various model tiers (Haiku, Opus).
• Quality Assurance: Participation in a rigorous, 4-layer human quality control process (creation, human review, calibration review, and audit) alongside automated LLM-based checks to ensure fairness, grammatical accuracy, and benchmark integrity.
• Linguistic Review: Reviewing AI benchmark tasks across Hindi, Arabic, Japanese, Chinese, Czech and Turkish.

• Setup of new desktop computers
• Setup of a VPS for automation, databases, and chat interface
• Connection to Azure AI Foundry
• Development of various n8n workflows for email, social media, and document management
• Implementation of a QM system (backups, error handling, HITL)
• Setup and optimization of an inbound Voice Agent

With project PRIZMA, DB will modernize its infrastructure on the one hand, and develop a fail-safe IT landscape on the other hand, which can be restored quickly and securely in case of a disaster.

• Capture current architectures of existing systems as well as methodical consulting and development of target architectures
• Deepen and maintain the building plan / target IT landscape
• Implement technical architecture concepts & architecture descriptions
• Implement migration concepts for updating and further developing the platform and information systems
• Assess submitted improvement suggestions as part of the project
• Capability management: identify capability gaps, develop target visions, and support transformation planning within the enterprise architecture.
• Create a compatibility matrix of the components in use and compare dependencies of specific versions
• Create an IT concept for extending the platform with the following topics: hardware and software requirements, security, licensing, high availability, load balancing, backup & recovery, update strategy, monitoring integration, etc.
• Coordinate with business architects as well as technical architects from the cross-functional architecture area of the PRISMA program for the topics (backup, Active Directory, monitoring, Citrix, and business applications ...)
• Status meetings and alignment of project planning with the Release Train Engineer / Project Manager
• Advise the Release Train Engineer / Project Manager in identifying project risks
• Advise the System Architect Engineers in steering the implementation of the concept
• Implement the IT concept
• Document the infrastructure

Label: MS Project, LINUX, Windows, ORACLE, Java, REST, SharePoint, Microsoft Exchange, UML, Enterprise Architect, BPMN, AZURE, AWS, V-MODEL, Micro Service, VisualStudio, SAP S/4HANA, SCRUM(SAFE), ESB (TIBCO), Python, Innovator, LeanIX (TOGAF), Ansible, Ansible Tower, Ansible Automation, ROBOT, SpringBoot

Workflow-automation SaaS for operations teams (Berlin, 120 people); full-time freelance engagement reporting to the CEO: an initial 12-month interim mandate, extended twice through the AI build-out; owned product for one squad and coached the other product managers on process.

• Led generative AI (LLM) integration into the core product: from LLM-powered steps to natural-language workflow authoring and step-level automation suggestions, plus AI-managed dynamic workflows, shipped behind eval gates with human-in-the-loop fallbacks: AI-drafted workflows grew to 31% of all new workflows, and median time-to-first-workflow fell from 3 days to 4 hours.
• Packaged the AI capabilities as a usage-based add-on priced on executed automation steps, working with sales and marketing on positioning: ~€800K added ARR in the first year, and adopting accounts churned 1.8 pp less.
• Owned the roadmap end to end: replaced feature-request-driven quarterly planning with an outcome-based rolling roadmap built on quarterly bets and explicit kill criteria, presented monthly to the executive team and quarterly to the board.
• Rebuilt the product-management operating system: weekly customer-discovery cadence incl. workshop facilitation, RFC/decision-doc reviews and a single quarterly metrics narrative; coached four product managers, one promoted to senior during the engagement.
• Closed the engagement as scoped: hired and onboarded the permanent VP Product, handed over the process playbook and roadmap, and exited on schedule in June 2026.

• Owned Qvantum, a cloud-based planning platform for personnel, cost center and sales planning
• Managed roadmap, prioritization and stakeholder communication
• Worked closely with development teams and business stakeholders
• Improved usability and supported data-driven decision making
• Led product development initiatives from discovery through delivery

Product development for the Miele app – design system, UX, and end-to-end product responsibility.

• Built and expanded the design system – components, design tokens, dev handoff
• Created user flows and customer journeys for core app features (device control, recipes, maintenance)
• Usability testing and iterative UX improvements
• AI-powered personalization – smart program recommendations based on usage behavior
• AI tools in the design workflow: automated accessibility checks, Figma AI, rapid prototyping
• AI-synthesized user research – condensed interviews, feedback, and usability data into structured insights

Scope: Embedded as FP&A Data & AI Architect within the Finance function of a major European refrigeration manufacturer, leading the transformation of manual, fragmented financial reporting into an automated, governance-driven intelligence platform. Driving the shift from Excel-based controlling to structured data architecture, Power BI analytics, and AI-assisted financial operations.

Financial Data Integrity & ERP Governance

• Initiated and led GL vs. subledger reconciliation investigations, identifying and resolving structural mismatches between General Ledger and subledger data that had gone undetected prior to engagement
• Conducted asset analysis to identify items missing from General Ledger postings, surfacing gaps in fixed asset tracking and period-end completeness
• Validated SAP reports, establishing baseline data quality standards for Finance team consumption
• Established systematic SAP data validation framework ensuring ongoing integrity between ERP postings and downstream reporting outputs

Finance Reporting Transformation

• Designed and implemented a structured Transformation Project approach for converting manual Finance reports into fully automated processes
• Created and owns the Data Reporting Audit Log; a centralized tracking system capturing report owners, stakeholders, data sources, manual effort estimates, and automation opportunity scores across the Finance function
• Mapped the full reporting landscape identifying quick-win automation targets and strategic Power BI migration candidates
• Actively reducing manual Excel and PowerPoint dependency across FP&A workflows; replacing point-in-time snapshots with live, governed data models

Power BI & Analytics Enablement

• Introduced and presented Power BI as the strategic reporting platform to Finance leadership, building internal buy-in for the BI transformation roadmap
• Designed initial Power BI architecture aligned with SAP, Salesforce and Oracle data structures and FP&A reporting requirements
• Established report ownership, governance documentation, and data lineage standards enabling sustainable self-service analytics across the Finance team

Transformation Infrastructure & Collaboration

• Configured and deployed Jira as the transformation project management hub, establishing structured sprint workflows, backlog management, and progress visibility for Finance IT initiatives
• Proposed and initiated a dedicated FP&A Communication & Transformation Hub, a structured cross-functional forum aligning Finance, IT, and business stakeholders around the reporting transformation roadmap
• Positioned the Finance function as an active driver of data governance and digital transformation within the broader organization

Outcomes

• GL/subledger reconciliation gaps identified and investigation framework established within first two weeks of engagement
• Data Reporting Audit Log deployed; first structured inventory of Finance reporting landscape in company history
• Power BI transformation roadmap presented and approved by Finance leadership
• Jira-based project governance live; Finance transformation now tracked with full sprint visibility

Technologies: SAP FI/CO · Power BI · DAX · SQL · Excel (advanced) · Power Query (M) · Power Automate · VBA · Jira · Microsoft 365 · SharePoint · Salesforce (Sales Data) · Oracle HCM · Python

• ▶ Built real-time voice assistant (STT → LLM → TTS pipeline) benchmarking and evaluating multiple STT providers including faster-whisper and Azure Speech. achieved sub-3s latency, Groq API (Llama 3) with multi-turn memory - directly handling edge cases in dictation, names and passcode recognition.