CloudCodeTree · Learn

Tutorials

Hands-on, hand-written guides to building and customizing AI — separate from the daily AI News feed.

Per page

Topics

RAG 7

Python 6

Fine-tuning 2

Search 2

Claude Code 1

tmux 1

Tailscale 1

SSH 1

Serving 1

vLLM 1

LoRA 1

Evaluation 1

Vectors 1

Embeddings 1

Claude Code Anywhere · Part 1/1

Your Whole Dev Machine, In Your Pocket

Self-host a free, secure pipeline from your phone to your real Mac with Tailscale, tmux, and one isolated Claude Code session per project. Your actual files and full toolchain — not a cloud sandbox.

Jun 25, 2026 · 12 min

Claude Code

tmux

Tailscale

Fine-Tuning & Serving · Part 3/3

Serve a Model with vLLM

Turn a fine-tuned model into a fast, OpenAI-compatible API. Merge the LoRA adapter, start a vLLM server with one command, and call it from the OpenAI client — a drop-in for your existing code.

Jun 24, 2026 · 8 min

Serving

vLLM

Fine-Tuning & Serving · Part 2/3

LoRA & QLoRA on One GPU

How parameter-efficient fine-tuning works: LoRA trains tiny adapter matrices instead of the whole model, and QLoRA adds 4-bit quantization so an 8B model fits a free Colab GPU. Concept + an official runnable notebook.

Jun 24, 2026 · 9 min

Fine-tuning

LoRA

Fine-Tuning & Serving · Part 1/3

Fine-Tuning vs RAG

The most common AI-engineering fork: retrieve knowledge (RAG) or change the model itself (fine-tuning)? A decision framework — when each wins, when to combine them, and what fine-tuning cannot do.

Jun 24, 2026 · 8 min

Fine-tuning

RAG

RAG from Scratch · Part 6/6

Evaluating Retrieval

Stop guessing whether a change helped. Build a golden query set, score retrieval with Hit@k and MRR, then A/B two retrievers over the same set — the capstone of the RAG track.

Jun 23, 2026 · 11 min

RAG

Evaluation

Python

RAG from Scratch · Part 5/6

Reranking

A bi-encoder retrieves fast but is fooled by negation; a cross-encoder reads query and passage together and fixes it. Retrieve wide, rerank narrow — with the recall ceiling made concrete.

Jun 23, 2026 · 10 min

RAG

Python

RAG from Scratch · Part 4/6

Hybrid Search

Vector search understands meaning; BM25 nails exact tokens. Each has a blind spot. Fuse them with Reciprocal Rank Fusion so retrieval stays correct when either one is wrong.

Jun 23, 2026 · 10 min

RAG

Python

RAG from Scratch · Part 3/6

Chunking Strategies

Chunking is the biggest lever on RAG quality. Compare whole-document, fixed-size, and paragraph-aware+overlap chunking on the same query — with real numbers.

Jun 23, 2026 · 9 min

RAG

Python

RAG from Scratch · Part 2/6

Give Your RAG a Vector Database

Move from in-memory numpy to Chroma: embed once, persist to disk, query fast, and filter by metadata. The next step after the RAG intro.

Jun 23, 2026 · 10 min

RAG

Vectors

Python

RAG from Scratch · Part 1/6

Build a RAG Over Your Blog

Your first hands-on AI-engineering project: retrieve answers from your own posts with embeddings + cosine search — no vector DB, no API key. ~60 lines of Python.

Jun 23, 2026 · 12 min

RAG

Embeddings

Python