Your second-biggest topic overall: 3,267 saved links. The recent corpus tells a clear story about your relationship with LLMs — it’s not model training or research, it’s operating LLMs as tools: Claude Code above everything else, then Copilot and Codex, understanding how models work under the hood, evaluating their output, and running them on Azure/Foundry. The theory-heavy material (papers, benchmarks) lives under ai-research; the harness/orchestration material lives in AI Agents.
Related: AI Agents · RAG · Prompt Engineering · Azure
Start here
- How I use LLMs as a staff engineer in 2026 — the workflow-level mental model.
- The Anatomy of an LLM — interactive explainer: text → tokens → vectors → attention.
- Hamel: the 11 links for learning AI evals — the copy-paste answer to “how do I learn evals?”
- Frontend Masters × Anthropic: free Claude Code course — no subscription required.
- Building Karpathy’s LLM Wiki: the complete guide — the idea this very vault is built on.
How LLMs actually work
- Transformer Explainer — interactive visualization of how GPT models work.
- How LLMs actually work — technical yet simple.
- Inside the Transformer: the life of a token — the deep-dive follow-up.
- Stanford CS336: Language Modeling from Scratch — full lectures + assignments.
- End-to-end LLM study material — components, quantization, pre-norm vs. post-norm, RL for LLMs.
- Alisa’s Book of LLMs + her math notes — OpenAI interview prep, free.
Claude Code
The single densest cluster in your entire corpus — this is your daily instrument, and your saves cover it from first steps to power use.
Learn it:
- Claude 101 → Claude Code 101 → Claude Code in Action — Anthropic’s own course ladder.
- Claude Code, clearly explained — the 15-minute onboarding.
- 100+ Claude Code resources, courses, tools and the claude-code-guide repo.
Master it:
- “I spent 6 months tuning Claude Code — the exact setup” — CLAUDE.md, subagents, hooks, skills, worktrees.
- My Claude Code setup — guardrails and a context/plan/code workflow.
- Stop prompting, start operating — rules, tests, skills, reviewers.
- Running Claude Code at scale — lessons from multi-million-line monorepos.
- The personal AI stack — teaching Claude your tools and conventions.
- Context control to cut your Claude bill.
- Architectural analysis of Claude Code and the primitive behind Claude Code, Codex, and Gemini — how these tools work inside.
- Deep Agents: build your own Claude Code — the batteries-included starting point.
Skills (the ecosystem your saves track obsessively):
The other coding assistants
- Copilot: Copilot CLI cheat sheet · the Copilot app · deterministic workflows with Copilot Hooks · GitHub’s own harness performance data.
- Codex: getting started with Codex (OpenAI Academy) · the free “Codex Orange Book” · maximizing Codex · how OpenAI uses Codex internally.
- Gemini: Managed Agents developer guide · the SRE extension for Gemini CLI — directly your lane.
Local & open models
- Run LLMs 100% free locally — the repo list (AnythingLLM, KoboldCpp & co).
- Unsloth: open LLMs inside Claude Code and Codex — Gemma 4 / Qwen3.6 GGUFs on 24GB RAM.
- Agents on local small language models.
- Hosting local LLMs in Docker on Azure — your two worlds again.
- Local-first inference with confidence-gate escalation to cloud — clever cost architecture.
Evaluation & quality
- Hamel’s 11-link evals curriculum (also in Start here — it’s the anchor).
- LLM Evaluation — the systematic overview.
- Teaching an LLM to review code like a senior engineer.
- Which SDLC stage eats the most tokens? (it’s code review).
- Best practices to secure code written with Claude.
LLMs on Azure & Microsoft Foundry
Your Azure depth meets your AI direction — the corpus is rich here.
- Microsoft Foundry: end-to-end workshop.
- Mastering Foundry IQ and migrating Azure OpenAI On Your Data to Foundry IQ.
- Azure AI Foundry anti-patterns — what not to do.
- Running Claude on Microsoft Foundry in your own tenant, with IaC — possibly the most you-shaped link in the corpus.
- Building Microsoft IQ: Work IQ + Fabric IQ + Foundry IQ.
LLM wikis & personal knowledge systems
The meta-section: links about the exact thing this vault is.
Industry & research signals
- Anthropic: how agentic coding amplifies some skills and substitutes others.
- What 81,000 people told Anthropic they want from AI.
- Diffing Claude’s system prompts between versions — Simon Willison’s forensic reading.
- “If not LLMs, what should I work on?” — finding blue-ocean problems.
- Fowler: let the LLM interview you for context — a technique, but a telling one.