Articles
Articles and insights
Lessons learned and technology watch, with a bias for technical explanations, diagrams and methods you can apply.
Structuring an AI-Assisted Development Harness: Maintainability, Security and Portability Across Assistants
A coding assistant is only as good as the environment it works in. A method for structuring an AI-assisted project: shared context, rules, tests, technical guardrails, data security and independence from any single vendor.
Read the articleReference Architecture for Sovereign AI: Five Layers, Flows, Threats and Evaluation
A reference architecture in five layers for controlled AI: Kubernetes and GPUs, an inference gateway, RAG with ACLs, constrained agents, identity, audit, threat model, continuous evaluation and reversibility.
Read the articleLocal AI: sizing memory, choosing an engine, deploying
Compute the memory of a model and its KV cache, pick a quantization, understand throughput and latency, compare llama.cpp, Ollama and vLLM, then deploy and monitor an inference server on your own machines.
Read the articleRAG in practice: making an LLM answer from your documents, with sources and access rights
Ingestion, chunking, embeddings, vector index, hybrid search, reranking, query-time access rights, prompt assembly and evaluation: the engineering guide to a RAG that cites its sources and stays under control.
Read the articleHow an LLM works: tokens, attention, sampling and limits
From the artificial neuron to the Transformer: tokenization, embeddings, attention, logits, temperature, context window, training, hallucinations, and what prompting, RAG and fine-tuning actually change. With calculations and code.
Read the articleData sovereignty in Africa and the Gulf: an architect's point of view
Residency, sovereignty and jurisdiction are not the same thing. Classification, data flows, key management, confidential computing, audit, reversibility: what an architect can really control, and what they cannot.
Read the article