September 7, 20267 min
GDPR and data residency once an LLM is in the loop
Where does the model run, where do embeddings and logs sit? The technical list that makes a privacy review of an AI system possible.
GDPRData residencyAICompliance
ReadBlog
Articles on cloud architecture, AI engineering and software development. From projects, not from a textbook.
September 7, 20267 min
The EU AI Act from the seat of the person building the system
Traceability, human oversight, documentation: what an engineer can contribute to the AI Act, and where the legal classification begins.
EU AI ActComplianceLLMRegulation
ReadSeptember 7, 20267 min
LLM operations: from demo to a run someone owns
A prototype needs an API key. Operations need versioned prompts, a pipeline, and someone who owns the run when it breaks.
LLMOperationsMLOpsDeployment
ReadSeptember 7, 20267 min
LLM cost: why the bill shows up after rollout
Context length, reranking and retries drive cost, not the model alone. What I measure per request before an LLM system goes live.
LLMCostFinOpsProduction
ReadSeptember 7, 202612 min
RAG in production
The path from a RAG demo to a system that survives an audit: chunking, hybrid retrieval, reranking, source grounding, evaluation, tenants.
RAGProductionLangChainEvaluation
ReadSeptember 4, 20268 min
Hybrid RAG and reranking instead of naive vector search
Character-count chunking and vector-only search make hallucinations expected. What hybrid search and reranking change, and what the older RAG article taught wrongly.
RAGHybrid searchRerankingMilvus
ReadSeptember 4, 20267 min
LLM evaluation in production
A prompt or model change makes quality worse. Nobody notices if evaluation is missing. What I measure in regulated LLM projects.
LLMEvaluationProductionRAGAS
ReadSeptember 4, 20267 min
A bad answer has to be reconstructable
Without observability nobody knows cost per request and nobody can reconstruct a bad answer. What production LLM systems must log.
LLMObservabilityLangfuseProduction
ReadSeptember 4, 20267 min
Tenant isolation for RAG and agents
A one-customer pilot is not a product for twenty. What tenant isolation means in RAG and agent systems.
RAGMulti-tenantAgentsSecurity
ReadAugust 28, 20268 min
LLM hallucinations are a compliance problem
When a model invents citations, legal stops the project. LLM traceability needs source grounding, not a better prompt.
LLMRAGComplianceSource grounding
ReadApril 10, 20268 min
Naive RAG: What chunk_size=1000 and vector-only search do
A healthcare project with LangChain, OpenAI and Milvus. The pipeline was fast. As a recipe it is dangerous: fixed chunks, vectors only, no reranking.
RAGLangChainMilvusOpenAI
ReadApril 5, 20267 min
Cloud Migration to Kubernetes: Lessons Learned from Government
A legal-communication application moves from WebLogic to Kubernetes. Four years, zero downtime, deployment from one day to 20 minutes.
KubernetesCloud MigrationTerraformDevOps
ReadMarch 28, 20267 min
MCP Servers: The Model Context Protocol in Practice
An MCP server for the healthcare chatbot: what the Model Context Protocol delivers in practice, and where tool descriptions and logging break.
MCPAILangChainPython
Read