Neural Lattice · Guides
The Reading Room
Concepts explained like you're a smart human who hasn't seen this before. No assumption you've been following along since 2022 — just clear, evergreen takes that don't expire with the news cycle. Bookmark these; the glossary is for definitions, this is for actual understanding.
What is RAG, in plain English?
Retrieval-augmented generation explained without the jargon — when you need it, when you don't, and common pitfalls.
ToolsHow to pick an LLM for your use case
Cloud vs. local, context length, cost per token, and latency — a decision tree for builders and architects.
SafetyAI agents: what can go wrong in production
Tool misuse, prompt injection, runaway loops, and audit trails — the guardrails checklist before you ship.
InfrastructureSelf-hosted LLM stack in 2026
A practical guide to running Ollama, Open WebUI, and MCP tools on a home NAS — privacy-first, no cloud lock-in.
BasicsContext engineering: the skill that replaced prompt engineering
Curating what a model sees — system prompt, retrieved docs, tool output, memory — not just what you ask it.
ToolsHow coding agents actually work
From autocomplete to autonomous: read, plan, edit, run tests, repeat — and where the loop still breaks.
MultimodalReal-time voice AI, explained
Speech-to-speech models replaced the old three-step pipeline — what changed, and where it still stumbles.
MultimodalVideo generation models, in plain English
How text-to-video actually works, and why it's still not final-pixel production ready.
ModelsReasoning models and test-time compute
Why "thinking" models spend extra compute before answering, and when that trade isn't worth it.
AgentsMulti-agent orchestration: when one agent isn't enough
Planner, worker, reviewer — common patterns for splitting agents, and when it's pure overhead.