Michael BoimanFreelance Quality Engineer & AI Architect · Frankfurt am Main

Profile

Michael Boiman

Freelance Quality Engineer & AI Architect · Frankfurt am Main

Pick a position, a competency or a project. The lines join each competency to the positions and projects that name it.

Available: on request
Based: Frankfurt am Main and remote
Travel: by arrangement
Billing: own account or BKS-Lab
Hourly rate: on request
Citizenship: German
References: on request

Quality engineer and AI architect based in Frankfurt am Main. Quality engineering and AI architecture run in parallel for me rather than in sequence: the testing discipline is the reason my agent system has gates and evidence instead of acting blind.

Try a question

Get in touch Download PDF

since 09/2026ongoing

AI Test Automation Engineer · Building the AI Test Automation Process

KION Group

AI-driven test automation for one of the group's customer portals: building the complete process, from deriving the test cases to a green CI pipeline.

Key Responsibilities

Test orchestrator: a chain of AI building blocks with a fixed log schema, its own model tier per block and usage tracking, packaged as a reusable command for the team
Architecture rules for the test code base: derived from a multi-stage analysis with several AI agents, independently cross-checked and documented
Architecture check for generated test code: detects violations of the team's style guide and fixes the minor ones on its own

Key Responsibilities · Tools & Technologies

Key Responsibilities

Calibrated comparison runs: AI-generated E2E tests compared on cost, usage and conformance to the standard, across several model tiers
Stable pipeline: found and fixed the cause of hanging tests (a third-party element was still missing while the page loaded)
Coverage: four test areas of a portal module end to end, from deriving the test cases to a green pipeline

Tools & Technologies

Testing: Playwright, TypeScript, Angular code base
AI: Claude Code (several model tiers), GitHub Copilot CLI
CI/CD & tooling: GitHub Actions, Sonar, Jira with Xray, Confluence, SAFe

since 06/2025ongoing

LLM Infrastructure Architect & Automation Engineer

BKS · AI Research & Development (internal)

Development of LLM-orchestrated Enterprise Development Ecosystem: Fully integrated system for knowledge management, project automation, and development workflows with Claude, Gemini, and OpenAI GPT.

Key Responsibilities

Open-Source Agent Infrastructure: open-bridge under the MIT licence, public at github.com/bks-lab/open-bridge, with a live demo site. The generic CORE layer (agent runtime, MCP gateway, skill tree, promote flow with a content-leak gate in CI) grows out of my own instance and is published upward by scope, with its own test suite and continuous releases since June 2026.
Harness engineering: what an agent achieves is decided mostly outside the model: context, tool selection, permissions, state across sessions, verification. That layer is measured here rather than estimated. What a session loads before it answers the first word carries a declared ceiling that a CI step checks in bytes; rules that rarely bite load on their trigger word instead of at start-up; the project registries sit in context as a card and are fetched one entry at a time; a further check reports what exists in the tree but is no longer reachable from any file a session loads, and attacks itself to prove it still catches things. The tools behind this are public and run in public CI
Data separation as architecture, not as agreement: separate instances per client, a rule forbidding access between instances, and a fail-closed contract for materialising shared configuration (the sync aborts rather than writing on doubt). Protection against data leakage at three points: pre-commit hook, pre-push hook and three validators in CI. Plus an adversarial simulation that attacks the protection itself on a schedule, as its own CI job in a sealed sandbox

Key Responsibilities · Tools & Technologies

Key Responsibilities

A2A Agent Network: an a2a-sdk server with a local claude backend, running on a runtime adopted from open-bridge. Three agents (person, company, OSS project) consult each other on read-only edges, fronted by a stateless MCP-to-A2A gateway so MCP-only clients can reach them. Agent Cards at protocol version 1.0, JSON-RPC 2.0. Around 4,600 lines of Python with 98 automated test cases, built as a team effort (four further contributors)
OpenTelemetry-based telemetry: logs, metrics and traces over OTLP into a managed sink, with a fixed field convention and naming scheme, a capped ingest key and read rights separated per client (measured as an access cut, not documented as an intention). Honest maturity: the log path is in operation, metrics and traces are wired and not yet accepted
Human gate on four autonomy layers: what an agent may change about itself is split into four layers (context, skill files and rules, user profile, model weights), each with an explicit human gate. Proposals go through a dedicated folder with a schema and a review path instead of taking effect silently. The EU AI Act is therefore not just workshop content here but my own operating practice
Specification and verification design: before building, a fixed sequence runs through requirement, design and verification plan. Every acceptance criterion is mapped to the strongest available check, in this order: deterministic test before external signal before model judgement, together with the gate strength. This is where quality engineering and AI development are the same discipline
Git Wiki Transformation: Migration of Confluence knowledge base to structured Git-based wiki with hierarchical organization and multi-repository architecture (submodules for customer projects)
MCP servers and protocol choice: Built two Model Context Protocol servers in Python, one for the wiki hierarchy (navigation, content discovery, SharePoint integration), one for the company website content. Delivered MCP training. In my own work, MCP is used only where it beats skills, REST or A2A, since most MCP servers are not effective in everyday use
Google ADK (earlier stage): the first version of the agent network ran on the Google Agent Development Kit with LiteLLM. Replaced in July 2026 by the a2a-sdk server on our own runtime; the ADK and tool-integration experience from that stage stays
GitHub Project Automation: LLM-driven issue management with 7-mandatory-field system, automatic categorization, smart repository mapping, and review queue management
Development Workflow Automation: GitHub Actions workflows for fully automatic issue-to-PR transformation (assignment → feature branch → implementation → code review → PR)
Self-Documenting System: LLM automatically writes project status, meeting protocols, and issue updates back to Git wiki; closed loop of wiki reading → work execution → results documentation → Git commit with structured logging
Work-time capture: commit-based work time analysis with automatic project assignment. The first stage ran against Clockify; since July 2026 it is being rebuilt as a private registry of semantic conventions on OpenTelemetry Weaver, with three layers (raw observation, derived block, billable time), canonical JSON serialisation per RFC 8785 and an integrity profile per OCSF
Claude Code Plugin Ecosystem: 120 skills in the productive instance, discovered alike by Claude Code, Codex, Copilot CLI and Cursor through committed symlinks: log analysis, deployment verification, incident handling, email management, meeting transcription (104 filed meetings), PDF/certificate generation, SharePoint integration. Every skill carries one of four scope tiers (core, org, personal, user) checked by a CI validator; what may be published outward depends on it
Context Orchestrator: central multi-repo control system with a three-way split registry (instance base, shared org fragment, personal fragment), so company data and personal data can be distributed separately. It carries repos, customers, infrastructure and skills, plus automatic cross-repo navigation and structured work tracking with archiving
Hook-based Permission System: Smart Guard with project-specific security rules for Git operations: branch protection, secret detection and deployment gates
Production Operations E-Invoicing: ongoing management of an e-invoicing platform for a leading online job platform (under NDA): automated incident management, log analysis, weekly reporting and go-live tracking
Peppol expansion of the same platform (from 06/2025): separate services for outgoing and incoming invoices (Azure Functions) next to the existing processing, connected to a certified Peppol access point; the SAP ByDesign interface built over SOAP instead of REST after a technical deep dive with the SAP partner; an extra inbound channel by mailbox for suppliers without Peppol (PDF, ZUGFeRD, XRechnung); operated through managed identity and Key Vault in separate pre-production and production environments
Self-Operated Infrastructure: 50 launchd services on the central host, 18 devices in the registry, reachable over Tailscale rather than open ports. Backups follow 3-2-1 as a declared topology: 10 sources, 5 targets, 16 pipelines (restic/rclone to NAS), with actual state reconciled against intent.

Tools & Technologies

LLM Orchestration: Claude (primary), Google Gemini, OpenAI GPT, Model Context Protocol (MCP), Google Agent Development Kit (ADK)
AI Development Platform: Claude Code, Plugin Architecture (Skills, Hooks, Commands, Agents), YAML/Markdown Configuration; Cursor before that
Multi-Agent: A2A Protocol 1.0, a2a-sdk, Agent Cards, MCP-to-A2A gateway, JSON-RPC 2.0, SSE Streaming
Backend: Python 3.13, FastAPI, uvx Distribution, Git Submodules
Automation: GitHub Actions, GitHub GraphQL API, gh CLI
Integration: SharePoint API, Microsoft Graph, Exchange Online, Elasticsearch, Kibana, OpenTelemetry/OTLP
Operations: Cloudflare (Workers, Pages, R2, DNS), Tailscale, launchd, restic/rclone
Cloud: AWS, Google Cloud (alongside Azure)
Development: Bash Scripting, jq, Docker, Multi-Agent Systems; orchestrated repos span Python, TypeScript, Rust and Go

09/2025-09/2026

Senior Quality Engineer · Test Automation and Monitoring

TÜV Süd

Quality Engineering for an enterprise platform for AI-powered document processing in medical device certification (MDR/IVDR) with AI chat integration and data pipeline automation.

Key Responsibilities

BDD API Test Framework: Development of comprehensive test framework with Python/Behave, self-healing authentication, HTML report generator with embedded API responses
E2E Test Automation: Playwright E2E test suite for complete user journey (Setup -> Upload -> Transformation -> Download) with 500 error handling and route mocking
Performance & Monitoring: Development of Azure Monitor Workbooks (QA Live Testing Companion, Error Investigation Assistant) with KQL for end-to-end monitoring

Key Responsibilities · Tools & Technologies

Key Responsibilities

Bug Analysis & Quality Intelligence: Implementation of comprehensive bug tracking system with WIQL queries, automated dashboards, and executive PDF reporting
Pipeline Integration: Test automation in Azure DevOps CI/CD pipelines with HTML report upload to Azure Test Results
AI-Powered Automation: Development of automation skills for workflow optimization, bug analysis, and incident response with measurable time savings
AI-Driven Test Processes: Integration, E2E and API testing set up as end-to-end processes with AI agents, plus AI-assisted incident analysis

Tools & Technologies

Testing: Behave, Playwright, TypeScript, Python, pytest, Page Object Model, Fixtures
Monitoring: Azure Monitor Workbooks, KQL, Application Insights, Log Analytics
Backend: FastAPI, Python, Pydantic, SQLModel, Azure Functions, Azure Service Bus
Frontend: React, TypeScript, Vite, TanStack Router
AI Integration: Azure OpenAI (GPT-4o), LangChain, Azure AI Search, Embeddings
DevOps: Azure DevOps Pipelines, Git, Docker, Azure Blob Storage, Poetry
Agentic Coding: GitHub Copilot CLI

07/2024-01/2025

AI-driven QA Environment for Energy Infrastructure

AkkuSwap Startup

QA Leadership for EU-wide Battery Swap Infrastructure: AI-driven simulation tool for battery swap station network with energy infrastructure integration

Key Responsibilities

QA concept, design and implementation for AI-driven simulation of EU-wide infrastructure
QA framework development for inhouse AI server infrastructure
Proof-of-concept for automated testing of AI-driven simulations

Key Achievements

Established QA methodology for AI-driven infrastructure simulation
Created testing framework for energy sector critical systems
Validated proof-of-concept with measurable reliability improvements

Energy Sector Connection · Technical Stack

Energy Sector Connection

50 Hertz Participation: 50 Hertz (Elia Group subsidiary) participated in research project eHaul, the predecessor of AkkuSwap
Direct energy sector experience with critical infrastructure quality requirements
Context of critical energy infrastructure (transmission-grid environment)

Technical Stack

Infrastructure: Linux, Docker, Azure OpenAI
Automation: pytest, AI code generation integration patterns
Monitoring: Grafana, Azure Monitoring
Energy Systems: Battery swap infrastructure, grid integration simulation

01/2024-04/2025

Technical Lead & AI Automation Architect

BKS on behalf of a client in the digital economy

Technical lead & development: Design and implementation of a scalable automation solution for processing e-invoices and PDF invoices fetched from a mailbox of a leading online job platform (under NDA). Full responsibility from requirements analysis through implementation to production go-live.

Focus areas

Workflow automation: Extraction, analysis and classification of incoming invoices using Python and Azure Functions.
E-invoice classification & data extraction: Full processing in accordance with EU standard EN 16931 (ZUGFeRD, XRechnung, Factur-X), including structural and content parsing.
Peppol Network Integration: Connection to European e-invoicing network through a certified Peppol Access Point for automated invoice receipt and dispatch with UBL/CII formats.

Focus areas · Tools & technologies

Focus areas

SAP ByDesign Integration: Automatic transfer of validated invoice data to SAP ByDesign for seamless ERP integration.
Hexagonal Architecture: Domain-driven design with Ports & Adapters for maximum testability and interchangeability of infrastructure components.
Salesforce integration: Forwarding prepared invoice data via email to the downstream system for Salesforce import.
Auditing & logging: Comprehensive documentation of all processing steps plus storage of original email content in the processing log.
Outlook management & SharePoint archiving: Automatic moving and categorising of emails into predefined Outlook folders; archiving of original and generated invoice documents via Microsoft Graph API.
Monitoring & logging: End-to-end real-time monitoring with Kibana, including alerting and dashboard creation.
Team coordination: Managing the developer team incl. code reviews.
AI Workflow Automation: Development of AI skills for automating task tracking, issue management, and project documentation
Intelligent Documentation System: Automated wiki synchronization with git commit analysis and natural language documentation generation
Automated Status Reporting: AI-driven email updates with automatic project progress analysis
MCP & A2A Protocol Implementation: Next-generation AI protocols for tool integration and multi-agent communication

Tools & technologies

Azure Functions: Serverless orchestration of automation workflows.
Python: Core logic for data extraction, classification and integration with automated CI/CD processes, packaging, and comprehensive unit and integration tests for quality assurance.
E-Invoicing: Peppol Network, Access-Point-API, ZUGFeRD/Factur-X, XRechnung, UBL 2.1, CII
ERP Integration: SAP ByDesign, Salesforce
Microsoft Graph API: Access to Outlook mailboxes and SharePoint libraries.
GitHub: Version control, CI/CD pipelines and collaborative development.
Kibana: Monitoring, log analysis and performance dashboards.
Streamlit: For developing an interactive web application for visualization.
AI Automation: Claude AI, Custom Skills Development, Git Analysis, Natural Language Processing
Protocols: Model Context Protocol (MCP), Agent-to-Agent (A2A), JSON-RPC 2.0

04/2023-04/2024

AI Engineering Lead & ML Solutions Architect

BKS on behalf of a publishing house

Project management & technical lead: Development and deployment of an AI-driven email bot to automate customer communication, AI-assisted processing of incoming emails and triggering of follow-up activities in existing SAP systems. Responsible for end-to-end project coordination and team management from idea to implementation.

Focus areas

Development of ML models: Design and training of specialized models for effective email classification, implemented via Azure Functions.
API integrations: Seamless connection to Microsoft Outlook 365 via Microsoft Graph API and to SAP systems for customer data checks.
Process automation: Automatic moving of emails based on classification and specific business rules, including complaint handling.

Focus areas · Tools & technologies

Focus areas

DevOps and monitoring: Implementation of CI/CD pipelines and use of Kibana for monitoring.
Development of a Streamlit-based user interface: Improved testability and visualization of process results.
Team coaching: Technical guidance for developers with regular code reviews and problem-solving sessions.

Tools & technologies

Azure Functions: For serverless application architectures and ML model execution.
Python: Main programming language for the email bot and ML models.
Microsoft Graph API: For accessing and processing emails in Microsoft Outlook 365.
SAP API: For querying customer data and communicating with SAP systems.
GitHub: For version control, CI/CD pipelines and collaborative development.
Kibana: For real-time monitoring and system performance analysis.
Streamlit: For developing an interactive web application for system monitoring.

08/2021-05/2025

Platform Quality Architect (CI/CD, Monitoring, Automation)

DVAG

Test management & coordination, as well as introduction of CI/CD with quality-gate setup. Acting as cross-team Quality Lead: onboarding, guidance, consulting and training for the introduction of quality metrics and test automation within the pipeline across teams.

Result: Real-time sprint-quality transparency via Grafana dashboards, quality-gated releases in the teams' CI/CD pipelines: data-driven release decisions replacing manual data gathering.

Key responsibilities

Design and implementation of functional and technical dashboards with Grafana for real-time quality measurement
Documentation of quality metrics covering the measurable sprint progress
Design and setup of CI/CD pipelines for multiple development teams (backend + frontend)

Key responsibilities · Tools

Key responsibilities

Introduction of quality gates to safeguard quality within the CI/CD pipeline
Adoption of Gauge (BDD)
Roll-out of Playwright and integration into the existing CI/CD pipeline
Functional and technical incident analysis
Consultancy, design and implementation of solutions for quality-related issues

Tools

IntelliJ / Visual Studio Code
Java / Kotlin / JavaScript / TypeScript
Playwright / Gauge / Karate
Python for processing and analysis of test and reporting data
CI/CD / GitHub (reusable workflows)
Helm / Containers
JIRA / Confluence
Grafana
Behaviour Driven Development
Consumer-Driven Contracts (Pact.io)

06/2021-08/2021

Founder: Boiman.Solutions and QualityCluster

Boiman.Solutions

Set up his own consultancy Boiman.Solutions and the quality engineering community QualityCluster, a group of quality engineers working towards more quality through a higher degree of automation.

01/2017-05/2021

DevOps Quality Lead & Dashboard Architect

DB Vertrieb

Test management, test coordination, test automation, agile (Kanban, Scrum, SAFe): Coordination as PO of a cross-team QA team including cross-team onboarding, support, consulting, training

Key responsibilities

Built the production load test for a bahn.de feature: load scenarios, test setup, execution and analysis
Load and performance tests within CI/CD with Gatling / JMeter, bottleneck analysis using Kibana, Graylog, Grafana and Instana
Agile quality assurance through the design of quality gates in the CI/CD pipeline

Key responsibilities · Tools

Key responsibilities

Design and implementation of functional and technical dashboards with Kibana, Graylog, Grafana, Instana & Elastic Search
Introduction of domain-specific logging for microservices to transparently track calculations and service interactions
Documentation of quality metrics on the progress of measurable sprint quality
Design and setup of CI/CD pipelines for data tests
Functional and technical problem analysis
Consulting, design and implementation of solutions for quality-related issues
Setup of Junit, unit tests and integration tests, among others with Java Cucumber
Development of a hybrid app for presenting (reviewing) results on a web or mobile platform
Coordination of integration tests between the involved teams

Tools

IntelliJ
Java
Cucumber / Cypress
JMeter / Gatling
CI/CD / GitLab CI / Jenkins
Helm / Docker
JIRA / Confluence
Kibana / Grafana / Instana / Graylog
AWS
Behaviour Driven Development
Consumer Driven Contracts (Spring Cloud Contract)
Ionic (hybrid app)
Selenium

09/2015-01/2017

Performance Engineering Lead & Test Architect

DB Systel

Load and performance tests and analysis (test manager, test designer, analyst): Development, execution and analysis of load and performance tests

Key responsibilities

Setup and execution of load and performance tests over HTTP, WS, REST, MQ and JMS
Analysis of results & performance logs
Documentation in the form of test concepts, analyses, management reporting and wiki documentation

Key responsibilities · Tools

Key responsibilities

Status reports
Customer consulting
Requirements management

Tools

Visual Studio
Eclipse
JMeter
SoapUI
Git
JIRA / Confluence
Excel & Access VBS
Interfaces: HTTP, WS, REST, MQ, JMS

04/2015-09/2015

Senior iOS Developer & Mobile Architect

Telekom

Senior iOS Developer: Further development of the rebuilt version of Deutsche Telekom's Customer Center App, extension with new features for IFA.

Key responsibilities

Swift
iOS 8
Requests & responses to backend

Key responsibilities · Tools

Key responsibilities

Error handling
iPad / iPhone
Scrum

Tools

Xcode
Git
JIRA

05/2009-03/2015

Quality Assurance Lead & Test Automation Architect

Siemens

QA, test management, test automation, requirements engineering: A total of 5 projects were supported as QA & test manager and successfully implemented

Projects

SOM (05/2009-07/2010): management and overview of OrgUnits; test management and requirements management
SIPCA (07/2010-09/2010 and 07/2014-03/2015): management system for employee annual targets and bonus calculation; change request management
STM (09/2010-03/2012 and 05/2014-07/2014): application and travel expense recording (Frontend: Web / Backend: SAP); test management and requirements management

Projects · Key responsibilities · Tools

Projects

PATAC (07/2012-08/2012): test automation
GPM2 (09/2012-05/2014): share bonus program for employees (Frontend: Web / Backend: JBoss); test automation

Key responsibilities

Creation of technical test concepts and test plans
Defect tracking
Management of CRs, communication between customers and developers
Creation, maintenance and scheduling of test automation
User acceptance tests
Training of new testers and test automation engineers
Support as test manager after project completion, in steady-state operation
Setup of acceptance tests
Transfer of functional specification and detailed specification to Quality Center
Monitoring of implemented requirements against the detailed design spec

Tools

HP ALM
HP UFT
HP QC
Quick Test Pro

03/2009-05/2009

Performance Test Engineer & Load Testing Specialist

ING-DIBA

Load and performance tester: The goal of the project is the design and development of LoadRunner scripts for load testing a bundle of approx. 20 J2EE applications in a banking portal

Key responsibilities

Creation of test scripts in LoadRunner with the http protocol
Execution of load tests and analysis of results
Monitoring and reporting

Tools

Tools

HP LoadRunner 8.1 and 9.5

01/2006-12/2008

Performance Test Engineer & Quality Consultant

British Telecom, Mobiliar, DB-Systel, Sparkassen Informatik, Loyalty Partner, Telekom, Itelium, Deutsche Post, Postbank

Load and performance tester / tester: Various projects, predominantly load & performance tests

Key responsibilities

Creation of test scripts in LoadRunner with the http protocol
Execution of load tests and analysis of results
Monitoring and reporting

Clients and systems

British Telecom (Munich): number porting system, load tests over HTTP and SOAP
Mobiliar (Bern): insurance system with Java applet and terminal front ends
DB Systel (Frankfurt): delay notification for rail passengers, load tests with simulated XML data
Sparkassen Informatik: banking portal of more than 30 J2EE applications on WebSphere (AIX and Sun)
Loyalty Partner (Munich): Siebel call centre application
Telekom (Darmstadt): document management system and a media system for a mobile operator
Itelium and Deutsche Post: load on point-of-sale systems
Postbank: transfer archive and client integration software

Tools

Tools

LoadRunner

04/2026 · Talks

Guest Lecture: AI-Assisted Software Development in Practice

TU Darmstadt · Information Systems Group · Lab course “AI Startup: From Idea to Execution”

Guest lecture in a block lab course for Bachelor's students in Computer Science (Dept. 20). Shared how agentic coding and AI prototyping work in real projects. The talk connects directly to the students' subsequent team work on their own startup ideas.

Topics

From prompt to prototype: AI toolchains that get an MVP done in hours, not weeks
Agentic coding in practice: where it scales, where the friction sits, where tools break down
Live demos from my own skill ecosystem, on demand in response to the group's questions
Lessons from production AI projects rather than slide theory

Format: moderated impulse talk with interactive live demonstrations. Audience: ~25 students with mixed prior knowledge.

06/2025 · Talks

Workshop: Development with Generative Language Models

Developer Workshop · Agent-based Software Development

Developer Workshop: Comprehensive workshop on modern AI agent development with practical demos and live coding

Workshop Content

Development with LLMs (Claude, Gemini, GPT-4o) and token optimization, used across projects
Model Context Protocol (MCP) & Agent-to-Agent (A2A) protocol implementation
Google Agent Development Kit (ADK) for multi-agent systems and tool integration

04/2025 · Talks

Presentation: Efficient Documentation through Automation

Enterprise Presentation · AI-powered Documentation Workflows

Enterprise Presentation: Demonstration of modern automation approaches for centralized, current documentation and sustainable project success

Key Focus Areas

Single Source of Truth: Central truth in Confluence, avoiding redundant versions
Shift-Left Documentation: Code and documentation developed in parallel, so both stay up to date
Automation First: Bots and CI/CD jobs for recurring tasks, automatic updates

2023 · Talks

Applications of AI in Business Context

AI Workshop for Business Integration

AI workshop: Comprehensive presentation and live demonstration of innovative AI applications for business with a focus on practical implementation, theory and code examples.

Presentation content

Comprehensive overview of AI fields and patterns (supervised/unsupervised/semi-supervised learning)
Deep dives into machine learning, deep learning and natural language processing
Detailed look at GPT models and how they work (next token prediction)

Peppol Network Integration: connection to the European e-invoicing network through a certified Access Point
Multi-Format Support: ZUGFeRD, Factur-X, XRechnung, UBL 2.1, CII (EN 16931 compliant)
ERP Integration: Automatic transfer to SAP ByDesign with validation and error handling
Hexagonal Architecture: Domain-driven design with Ports & Adapters, seven repositories
More test code than production code: around 46,700 lines of application code against 81,600 lines of tests, roughly 2,900 automated test cases. Every one of the seven repositories enforces a minimum test coverage in CI, 75 percent in the core service
Format coverage: ZUGFeRD and Factur-X from 2.0.1, XRechnung in UBL and CII syntax, UBL 2.1, PDF/A-3 with embedded XML, plus a format detector that tells the variants apart by namespace
Dual-Channel Inbound: access-point webhook (push) + email inbox (pull) for redundant processing

Impact: Production system: Peppol connectivity per EN 16931, automated inbound and outbound invoice processing through to the ERP. The outbound route runs 13 processing steps from the SAP document to Peppol delivery, each independently testable.

Python · Azure Functions · Peppol · Peppol Access Point · SAP ByDesign · ZUGFeRD · Factur-X · UBL · Hexagonal Architecture

Quality Engineering

24/7 Automated Legacy Migration Validator: Dual-Run Quality Engineering

Enterprise Legacy Transformation with End-to-End Quality Assurance

The Challenge: Complex mainframe price calculation with millions of combination possibilities needed secure migration, without quality loss or data errors.

The Solution: 24/7 Dual-Run Validation System with intelligent deviation analysis

Continuous Testing: Randomly generated tests across all variations around the clock
Real-time Monitoring: Live dashboard with deviation detection on every deployment
Complete Traceability: Every deviation uniquely logged with technical and business context
Automated Root Cause Analysis: Direct access to calculation paths and error sources
Quality Intelligence: Kibana dashboards for end-to-end quality overview

Result: Secure migration without data loss, 24/7 validation across all combinations and full transparency over migration status

Java · Cucumber · Kibana · Graylog · Elasticsearch · Container · CI/CD · Mainframe · Dashboard Engineering · Test Automation

2025 · AI & Machine Learning

Agent mesh and MCP-to-A2A gateway

Three agents, one translator, one open protocol. The agent on this page is one of them.

Problem: An agent that only holds its own knowledge guesses at everything outside it. Loading it up with someone else's knowledge does not make it smarter, only more expensive and harder to check.
Solution: Three agents, each with its own subject, a person, a company and an open-source project, consult each other over the open A2A protocol, read-only. In front of them sits a stateless gateway that offers the same agents to MCP clients as three tools, with no model and no logic of its own.
Result: A subject question gets passed on instead of guessed, and the borrowed answer is marked as borrowed. The agent card is publicly retrievable, and the generic framework behind it is open under MIT.

Python · a2a-sdk · A2A Protocol 1.0 · Agent Cards · JSON-RPC 2.0 · SSE Streaming · MCP · FastAPI

bks-lab.github.io/open-bridge/agents.html#gateway

Quality Engineering

Quality Dashboard · Real-time Overview

End-to-end solution for real-time quality monitoring: the test tools of several teams and environments under one surface, plus multi-source integration (Pipeline, LeanIX, Jira, Confluence) for automated PDF reports with open issues, bugs, and technical debt.

Grafana · Elasticsearch · Python · CI/CD · Playwright · Jira API · Confluence API · LeanIX · PDF Generation

2024 · AI & Quality Engineering

AI-driven Test Automation

Guideline-based AI agent in the IDE that generates tests directly from Jira stories.

Problem: Deriving test cases from a story is manual work, and it comes back with every story. A language model without constraints invents steps and selectors the application does not have.
Solution: The agent reads the story over the Jira API and writes the test from it. The project's own conventions, page objects, naming rules and permitted selectors, sit beside it as a rule file and bound what it may produce.
Result: The test comes out in the style of the existing Playwright project. Typing an empty file turns into reviewing a draft.

Python · OpenAI GPT-4 · Playwright · IDE Integration · Jira API

2025 · AI & DevOps

AI Context Orchestrator: Multi-Repo Development Platform

Central Control System for Enterprise AI Development

Core Features:

Context Orchestrator: a three-way split registry (instance base, shared org fragment, personal fragment) for repos, customers, infrastructure and skills, so company data and personal data can be distributed separately
One skill tree for every tool: discovered alike by Claude Code, Codex, Copilot CLI and Cursor through committed symlinks. Log analysis, deployment verification, incident handling, email management, meeting transcription, PDF/certificate generation, SharePoint integration
Hook-based Permission System: Smart Guard with project-specific security rules (branch protection, secret detection, deployment gates)
Structured Work Tracking: Automatic documentation with daily/weekly archiving and cross-project Kanban board
Cross-Repo Automation: Seamless navigation and operations across customer, partner and internal repositories
Open-source layer: three-tier architecture. The generic CORE layer is public as bks-lab/open-bridge (MIT, with a live demo site) and is fed from this seed instance, with a BKS overlay in between.

Impact: In daily use: 120 skills, 5 sub-agents, 15 standing orders, 104 automatically filed meetings. Six mutually isolated instances share the same open-source core without data flowing between them.

Python · YAML · Claude Code Plugins · Hooks API · GitHub Actions · Elasticsearch · Azure Functions · Microsoft Graph API

github.com/bks-lab/open-bridge

2025 · Training & Workshops

Developer Workshop: AI-Agent Development & Integration

Target Audience: Developers, Technical Leads, DevOps Engineers

Workshop Content:

Hands-on AI Agent Development: Live coding with Claude, Gemini, GPT-4o
Next-Gen Protocols: MCP & A2A implementation for enterprise integration
Browser Automation: Playwright-MCP for automated test generation
IDE Integration: AI assistance workflows with Claude Code as primary stack, GitHub Copilot & VS Code as tool comparison
Multi-Agent Orchestration: JSON-RPC, secure agent communication
Practical Demos: Jira integration, Elasticsearch queries, GitHub workflows

Deliverables: Complete source code, slides, hands-on exercises, production-ready templates

Anthropic Claude · Google Gemini · OpenAI GPT-4o · MCP · A2A · Playwright · JSON-RPC · IDE Integration · CI/CD

Consulting & Strategy

Enterprise AI Consulting: From Strategy to Implementation

Target Audience: C-Level, Management, Product Owners, Business Analysts

Consulting Services:

Strategy Development: From AI vision to measurable implementation roadmap
Process Analysis: Identification of automation potential, quality-first assessment
ROI Assessment: Business cases, cost savings, amortization calculations
Executive Workshops: Hands-on strategy sessions with live-demo business cases
Change Management: Team adoption, training concepts, resistance management
Implementation Support: From concept to productive AI solution

Deliverables: AI strategy template, ROI calculator, implementation roadmap, executive workshop, implementation support

AI Strategy · Process Analysis · ROI Assessment · Implementation Planning · Change Management · Executive Workshops · Quality Engineering

2025 · Healthcare AI

Medical Transcription System

GDPR-Compliant Transcription of Doctor-Patient Conversations

Core Features:

Fully local: No cloud uploads, all processing on local machine (Privacy-by-Design)
Speaker Diarization: Automatic doctor/patient distinction with Pyannote 3.1
SOAP Protocol Generation: Automatic creation of structured medical protocols
Open Source AI: Ollama with Llama 3.2 for intelligent summaries without API costs
Apple Silicon Optimized: Native performance on M1/M2/M3 Macs

Impact: Privacy by design, with fully local processing (no cloud upload), automatic speaker separation and structured SOAP protocols.

Python · Whisper · Pyannote · Ollama · Llama 3.2 · Apple Silicon · SOAP Protocol

2025 · E-Learning & Compliance

AI Learning Platform

EU AI Act Compliant Learning Platform for AI Workshops

Core Features:

Article 4 Compliance: EU AI Act compliant certifications with audit trail
Learning Card Types: Interactive quizzes, exercises, adaptive tests with scoring system
Visual Themes: Learning, AI-Team, Star Trek, Nerd, Helldivers, Wednesday
Progress Tracking: localStorage-based tracking with HR export functionality
Automated Tests: learning paths and certification logic are covered by regression tests

Impact: Around 18,400 lines of application code and 16,100 lines of course material across 524 versioned files, written single-handedly (116 of 117 commits, the remaining one from a bot).

Astro · React 19 · TypeScript · Supabase · EU AI Act

2023 · Quality Engineering

Django SLA Dashboard · Enterprise Service Monitoring

Enterprise SLA Monitoring Dashboard for Service Level Agreements

Development of a comprehensive SLA monitoring system as part of a BKS evaluation project (2023-2025) for monitoring Service Level Agreements with real-time metrics and automated alerting.

Core Features:

Service Registry: monitored services, each with its own SLA target and check interval
Response Time Capture: periodic measurement per service, stored as a time series
SLA Evaluation: availability against the configured target, rolling over selectable windows
Incident Management: capture and prioritisation of open incidents
Performance Analytics: response time distribution as a histogram over configurable buckets
Availability Timeline: history over a selectable window at a chosen resolution, with trend display
Alert System: rule-based alerting on SLA violations and performance degradation

Technical Highlights:

Django 4.2 with async views for high-performance data processing
PostgreSQL with TimescaleDB extension for time-series data
Django Background Tasks for asynchronous processing
WebSocket integration for live dashboard updates
RESTful API for service integration and CSV export
Docker-based deployment with Kubernetes-ready architecture

Context: Internal evaluation and demonstration dashboard fed by simulated SLA data, an architecture reference for time-series monitoring (not a production system).

Django · Python · PostgreSQL · Chart.js · WebSockets · Docker · REST API

github.com/bks-lab/experimental-django-sla-dashboard

2021 · Quality Engineering

QualityCluster

QualityCluster is an alliance of quality engineers jointly developing better solutions in software quality assurance.
Our goal is to implement more quality through a higher degree of automation.

Hugo · JavaScript · GitHub Actions · CI/CD

2024 · AI & Machine Learning

Streamlit demos: NLP text analysis

Interactive Streamlit demonstrations for NLP text analysis and AI-driven document processing.

Streamlit · Python · NLP · spaCy · NLTK · PyPDF2 · Named Entity Recognition · Sentiment Analysis · OpenAI

2021 · Mobile & Web Development

wecation

wecation is a collaborative solution for organising shared experiences: plan, search, coordinate with friends or colleagues and enjoy the activity together! Built with Flutter and Google Cloud.

Flutter · Google Cloud · Firebase · Dart

wecation.de

2023 · AI & Machine Learning

AI-Driven Chatbot for Trade-Fair Interaction

Development of an interactive chatbot using RAG (Retrieval-Augmented Generation), OpenAI GPT-4, Vector Embeddings, Python FastAPI, and Streamlit frontend to optimize user interaction on trade-fair websites.

Python · OpenAI GPT-4 · Streamlit · FastAPI · Vector Embeddings

2026 · AI & Machine Learning

Appointment enquiry over WhatsApp and phone

Two channels on one domain core: a WhatsApp assistant and a phone assistant that read free slots live from the booking portal instead of inventing them.

Problem: Calls were lost during working hours, and a language model must never invent an appointment.
Solution: The model runs the conversation, the free slots come live from the portal API. It pencils an appointment in but never books it, and consent comes before the first substantive sentence.
Result: The WhatsApp channel has been in production since August 2026 on the client's own business account, with data residency in Germany and real end users. The voice channel is built and verified and is deliberately idle at present.

Python · FastAPI · WhatsApp Business Cloud API · Speech-to-Text · Text-to-Speech · LLM · Telephony gateway · Booking portal API

2023 · AI & Machine Learning

Email Classification and Processing Workflow

Complete automation solution with Microsoft Graph API, SAP RFC, Azure Functions (Python), OpenAI GPT-3.5-Turbo, Elasticsearch 8.x, Kibana Dashboards, Docker containers, and GitHub Actions CI/CD.

Python · Azure Functions · OpenAI GPT-3.5 · Microsoft Graph API · Elasticsearch · Kibana · Docker

Swipe sideways to explore