AI Engineering
Building with LLMs: integration, RAG, evaluation, cost control, and product patterns.
LLM integration in SaaS: architecture patterns that survive production
Building language models robustly into SaaS: gateway layer, tenant context, structured outputs, evaluation, cost per tenant, fallbacks and data residency.
Retrieval-Augmented Generation (RAG): A Practical Engineering Guide
RAG in practice: preparing documents, chunking sensibly, hybrid search, reranking, enforcing permissions, showing sources and making quality measurable — with notes on German-language text.
AI Cost Management: Keeping LLM Spend Predictable in SaaS
How SaaS teams control LLM costs: per-workspace metering, quotas, model choice per task, caching and pricing models that keep an AI feature profitable instead of eating the margin.
Prompt Engineering for Products (Not Demos)
Prompts in a product are code: versioned, tested, structured. How to build system prompts, enforce outputs, separate inputs from instructions and keep prompts maintainable.
Evaluating LLM Features: Testing What Has No Single Right Answer
How to test AI features systematically: evaluation sets from real requests, checkable criteria, models as judges with human checks, and regression tests on prompt and model changes.
pgvector vs. Dedicated Vector Databases: Choosing for SaaS Scale
When pgvector in PostgreSQL is enough for SaaS and when a dedicated vector database pays off: tenant isolation, filters, indexes, operations and the questions that decide the choice.
AI Agents in a SaaS Product: Where They Help and Where They Hurt
When AI agents add real value in SaaS and when they create risk: levels of autonomy, narrowly scoped tools, approvals, prompt injection, logging and cost control.
Claude API vs. OpenAI for EU SaaS: Data Residency and Practical Trade-offs
Anthropic Claude or OpenAI for a European SaaS product? A neutral comparison of the criteria: processing location, contract terms, access via cloud platforms, model choice and dependency.
Designing AI Feature UX: Trust, Latency and the Empty State
How to design AI features users trust: set expectations, bridge waiting time, make results editable, show sources, communicate errors honestly and label AI.
AI Features and GDPR: Lawful Processing When an LLM Touches Personal Data
What SaaS providers must settle when AI features process personal data: legal basis, processor agreements, international transfers, data minimisation, transparency and data subject rights.
Fine-Tuning vs. Prompting vs. RAG: Choosing the Right Tool
When is a good prompt enough, when do you need RAG, and when does fine-tuning pay off? A decision guide by type of problem: knowledge, behaviour, format, cost and freshness.
EU AI Act and General-Purpose AI: What LLM-Based Developers Need to Know
Which AI Act obligations apply to model providers and which to companies building GPAI models into their products? Roles, deadlines, information duties, fine-tuning and transparency.
AI Observability: Logging, Tracing and Detecting Quality Drift
What to watch in AI features in production: traces across multi-step flows, latency, cost, errors, quality signals and drift — logged in a privacy-compliant way.
AI Content Moderation in SaaS: Safety Without a Trust-and-Safety Team
How small SaaS teams moderate user-generated content with AI: policy first, tiered review, human decisions on edge cases, statements of reasons, appeals and the obligations under the DSA.
AI-Driven Onboarding: Personalising the First Five Minutes
How AI can improve SaaS onboarding without being intrusive: ask for the goal instead of guessing a profile, prepare starter content, simplify imports, offer contextual help — and where personalisation backfires.