theLAB / Practical learning
Don't just read it.
Build it.
Twelve field guides. Three learning paths. Take an idea from a blank terminal to a working system, with checkpoints, failure modes, and a way back.
Companion to the wiki: explanations live there; step-by-step builds live here. Reviewed October 27, 2026. Publication checks are not a claim that every hardware build has been run end to end.
Path 01
Your own AI
Start with a working local model, add a chat workspace, then make it recoverable.
Your First Local AI Chat Server
Run a local model, send your first API request, and check where inference happens.
Build a Private Chat Workspace with LibreChat
Connect a persistent chat workspace to your local model without publishing its databases.
Back Up Your AI Workspace and Restore It
Capture a consistent recovery set and prove it works in a separate environment.
Path 02
Make it useful
Give your assistant documents, tools, a voice, and source-linked transcripts.
Ask Questions About Your Own Documents: Build a RAG System
Build a complete small RAG pipeline: documents, chunks, local embeddings, retrieval, and cited answers.
Build Your First MCP Tool
Build a small SQLite notes server and connect it to a tool-capable assistant.
Build a Local Voice Assistant
Record a request, transcribe it, ask a local model, and speak its response.
Turn Video and Audio into Searchable Notes
Choose captions or local ASR, preserve timestamps, and create source-linked Markdown notes.
Path 03
Build and operate
Develop, compare, publish, and troubleshoot with clear boundaries.
Run a Coding Agent in Your Browser
Give an authenticated browser agent a disposable repository and verify its change.
Choose the Best Local Model for Your Hardware
Compare task quality, warm and cold latency, generation speed, and memory use.
Connect Two AI Servers with Safe Failover
Route fresh requests to a healthy local backend and test failure without cloud fallback.
Publish a Self-Hosted App Safely
Expose one harmless web app over HTTPS with an explicit access policy and private origin.
Diagnose a Broken Docker Application
Repair a deliberately broken disposable app by tracing DNS, ports, and networks.