Ndamulelo Nemakhavhani · AI · Cloud · Data
Field notes from building with artificial intelligence, cloud platforms, data systems, and the software connecting them.
Latest field note
Overlapping plugins complicate attributing coding-agent outcomes to the model or its configuration. This post distinguishes package resolution from per-run behaviour and proposes controlled comparisons using invocation traces, correctness, safety and cost.
Read the essay →Code repositories, documents, and agent instructions answer different questions about a software system. This guide maps generated repository wikis, codebase analysis, RAG, linked knowledge wikis, and the Open Knowledge Format to help teams choose the right documentation architecture.
The MCP authorization draft deprecates Dynamic Client Registration as a compatibility fallback, not as an OAuth mechanism globally. Client ID Metadata Documents offer portable, origin-associated metadata, but do not authenticate a running public client. This article explains the operational trade-offs, localhost impersonation risks and metadata-fetch protections that remain.
An OAuth proxy can connect MCP clients to an Entra-backed service while keeping downstream registration separate from upstream application provisioning. This conceptual walkthrough explains the two authorization transactions, per-client consent, token validation and refresh limits that make the proxy a security-critical authorization server.
Your UI is just one client now; your apps must now treat humans, LLMs, and peer agents as equal consumers.
Azure Monitor quietly shipped a native OTLP endpoint. Here's how to send traces, metrics, and logs to App Insights from a plain Python app — no vendor SDK, your instrumentation stays portable.
I led the redesign and migration of a VM-based deduplication solution to Azure Batch. Moving the workload to managed batch compute improved scalability and reduced annual client infrastructure costs by approximately $8,000.
I owned the research and training of masked language models for Tshivenda and other indigenous Bantu languages. The work expanded language-technology coverage for languages underserved by mainstream NLP systems.
I applied NLP and large language models to extract key information from regulatory documents. The workflow was designed to make review more structured, traceable, and less dependent on repetitive manual handling.
