Day 34 primary goal: complete the How AI Actually Works — From Silicon to Response series. All 7 parts written, styled consistently, and validated clean. The series gives BU ITS staff a guided narrative walkthrough of the complete AI stack — from hardware through training, the LLM engine, inference, interface, and applications.
The TMA pages written in October 2025 now have a front door and a narrative thread connecting them.
ai-stack-index.html — series landing page, 7-card grid, TMA reference linksindex.html → ai-stack-index.html — symlink in place01-ai-stack-overview.html — The AI Stack Overview, six-layer diagram, sysadmin web-app analogy02-hardware-layer.html — GPUs, TPUs, clusters, VRAM, inference vs training hardware03-data-and-training.html — data pipeline, pre-training, fine-tuning, RLHF, frozen model04-llm-engine.html — transformer, tokenization, self-attention, parameters, context window, temperature05-inference.html — inference pipeline, forward pass, autoregressive generation, streaming, batching, KV cache06-interface-layer.html — API, web UI, CLI tools, system prompt, prompt anatomy, prompt engineering07-applications.html — application categories, RAG, agents, sysadmin use cases, full stack recapglossary/ directory — 2 files, vnu cleanEvery page follows the same pattern: green intro box framing the problem, sysadmin analogies in green callout boxes, content sections, TMA deep-dive links, and series prev/next navigation. The audience is BU ITS staff — technically literate, not AI specialists. Analogies to data centers, log files, pipelines, config files, and shell scripting carry the concepts without requiring a machine learning background.
Part 7 closes with the full six-layer stack reproduced and a completion statement: "From silicon to response — that is how AI actually works."
AI-NEW.html — add entry point card for how-ai-works/TMA/AI/) and Template F (TMA/LLM/)collaboration/lessons-learned.html — Days 11–34 unrecordedDay 34 complete • July 8, 2026 • Craig & Claude collaboration