Siegfried-Thor Bolz

Adobe Experience Manager has been my home CMS for 15+ years — and it is still the core of what I do. I architect, optimise and migrate enterprise AEM platforms end to end, from AEM 6.5 OnPrem to AEM as a Cloud Service (AEMaaCS), staying deep in the parts that make a CMS genuinely hard: the JCR content repository, Sling Models, Core Components, Editable Templates, Experience Fragments, Context-Aware Configurations and complex authoring workflows.

My current flagship is an AI-first enterprise CMS on AEMaaCS — migrating tens of thousands of pages and large media archives out of heterogeneous legacy systems into one unified cloud platform, with an intelligent DAM, an in-editor AI co-pilot, agentic content workflows and automatic multilingual translation. I have also led CoreMedia → AEMaaCS migrations for a leading European telecommunications provider, designing complex B2C, B2B and B2P customer journeys.

When AEM is the wrong fit, I map AEM Exit strategies toward composable, headless architectures with Contentful or Storyblok, Next.js / Nuxt and microservices on Azure Container Apps or Google Cloud Run. And because I have hardened AEM for government and telecommunications clients, I now extend that platform depth into AI risk auditing and cybersecurity — the red thread of my offer: from a rock-solid CMS to a governed, secure, end-to-end AI integration.

My AI risk audits cover AI embedded in enterprise CMS platforms as well as existing LLM and agentic systems. I map use cases, model and provider dependencies, data flows, tool permissions, autonomy and human oversight. I set audit depth in proportion to potential harm, data sensitivity and operational impact; EU AI Act classification is a separate step.

I independently review application code and architecture and compare policies, contracts and product claims with the controls that actually enforce them. For RAG and agents, I examine untrusted content, prompt-injection paths, retrieval and tool authorisation, evaluation, monitoring and change control. Depending on scope, the NIST AI RMF, ISO/IEC 42001 and ISO/IEC 23894 structure the review.

Clients receive an evidence-based readiness and gap assessment: verified strengths, prioritised findings, actionable remediation and clear owners. I distinguish engineering fixes from decisions for management. A system inventory, control matrix and living risk register make follow-up audits traceable; model cards and an AI-SBOM are added when they fit the system.

Content is the perfect fuel for Retrieval-Augmented Generation (RAG) — and a CMS is where that content already lives, structured and governed. I design and review GenAI and RAG systems on top of content platforms: LLM integration, vector search over CMS content, semantic retrieval and agentic workflows that turn a repository into an answer engine.

As an AI Backend Engineer & RAG Architect I have built production pipelines on Google Cloud: LangChain and FastAPI microservices on Cloud Run, embeddings on Vertex AI (Matching Engine, Pinecone), Gemini 2.5 Pro, Redis (Memorystore) caching and BigQuery analytics. Real deliverables include a RAG-powered intranet chatbot, an in-editor AI co-pilot, and AI product recommendations linking AEM to a SAP PIM system via semantic search. For new projects, I also design RAG architectures on Azure with Azure AI Search and Azure OpenAI.

Building these systems also informs how I audit them. I independently trace data, retrieval, tool and authorisation paths in existing RAG and agentic applications, then assess prompt injection, poisoning, confused-deputy behaviour, memory manipulation and tool misuse against the OWASP Top 10 for LLM Applications, the OWASP Top 10 for Agentic Applications and MITRE ATLAS. I distinguish code-review findings from hypotheses that need authorised runtime tests.

For RAG and agentic AI reviews, I use OpenAI Daybreak Blue and Claude through my approved access to the Anthropic Cyber Verification Program. For authorised defensive analysis, I examine prompt-injection and access-control findings, trace affected code and data flows, and develop and validate mitigations. I ground the conclusions in testable evidence.

I design and build secure, auditable RAG environments on Azure that connect governed enterprise content to retrieval and language models. I use my Auditable Modular RAG as the reference architecture, adapting its versioned module contracts, protected access and corpus eligibility, source provenance and evidence-backed release gates to each use case. On Azure, the design connects content ingestion, Azure AI Search and Azure OpenAI in Microsoft Foundry.

I also audit existing Azure RAG systems using this reference as a review framework. I trace content and permission flows, index and publication lifecycle, retrieval quality, prompt-injection exposure and output controls. I verify whether safeguards are enforced and supported by tests and operational evidence, then deliver a traceable assessment of gaps and improvements.

I integrate Contentful and Storyblok as SaaS headless CMS platforms into existing enterprise architectures. For each platform and use case, I design structured content models, localization, editorial workflows, API delivery and integration with surrounding systems. The chosen CMS remains the editorial source for its content; application identity, access decisions and business data stay with the systems responsible for them.

I also design Contentful- and Storyblok-to-RAG integrations: platform-specific adapters validate published content and make it retrievable in a RAG system, for example on Azure or Google Cloud. Publication changes and withdrawals, access rules and source links remain part of the design, so answers use permitted, current content and can be checked against the source.

Enterprise CMS platforms are high-value attack surfaces — public-facing, content-rich and heavily integrated — and AI only widens that surface. I build security in from the architecture stage: OWASP Top 10 coverage, hardened content and headless APIs, secure authoring and publishing flows, threat modelling and secure-by-design patterns instead of fixes bolted on after launch.

This comes from 15+ years of secure enterprise Java and AEM for government and telecommunications clients — integrating security code scanners, remediating real vulnerabilities (for example HTTP Method Override), keeping Maven dependencies patched, and securing API routing with FastAPI gateway validation and IAM. My working toolchain: Burp Suite, OWASP ZAP, Snyk and SonarQube.

For AI-enabled platforms, I independently trace how external content, agents and tools cross trust boundaries and whether safeguards are enforced in code. I document effective controls alongside vulnerabilities, prioritise fixes by realistic impact and use authorised testing when the scope calls for it. This connects adversarial AI, LLM security and secure software engineering from design through remediation.

For authorised defensive work, I use OpenAI Daybreak Blue and Claude through my approved access to the Anthropic Cyber Verification Program. I analyse and prioritise security findings, review affected code, work through remediations and validate patches. I check the evidence and proposed fixes before recommending them.