Raspberry Pi AI Projects Book Covers Local LLMs for £9

An Official On-Ramp to Running AI Without the Cloud

Raspberry Pi Press published *AI Projects with Raspberry Pi* on July 21, 2026, and the reason it is worth flagging is coverage rather than novelty. Local AI on a single board computer has been achievable for a while; what has been missing is one place that walks through the whole surface — vision, speech, transcription, translation, custom model training, and running large language models locally — on hardware people already own.

• Covers eight project areas including computer vision, speech-to-text, text-to-speech, transcription, translation, sensor-data ML, local LLMs, and image generation with Stable Diffusion
• Projects span Raspberry Pi 4, Pi 5, Pi Zero 2 W and the Pico microcontroller
• Uses the Raspberry Pi AI Camera, AI HAT+, and the newer AI HAT+ 2 accelerator
• Launched at an introductory £8.99, down from £17.99, in print plus PDF and ePUB

What Hardware Do You Actually Need?

That is the question these projects answer better than most tutorials. The book spreads across the full range — Pico for the microcontroller-class work, Pi Zero 2 W for low-power always-on builds, and Pi 4 or Pi 5 for anything involving a language model. The accelerator accessories are where the interesting scaling happens: the AI Camera handles inference on-sensor, while the AI HAT+ and AI HAT+ 2 add dedicated NPU throughput to a Pi 5 for vision and detection workloads.

Local LLMs get their own treatment, which is the section most people will turn to first. Running a quantized model entirely on a Pi is not going to match a datacenter endpoint, but it does something a datacenter endpoint cannot: it keeps everything on your desk, with no API key, no per-token bill, and no network dependency.

The Training Chapter Is the Underrated One

Alongside deployment, the book covers training custom models — including a "magic wand" hand-gesture recognition build driven by sensor data. Gesture recognition from an accelerometer is a genuinely good first training project: the dataset is small enough to collect in an afternoon, the model is small enough to run on a microcontroller, and the result is immediately, physically obvious when it works.

Where It Fits Alongside the Hardware

The accessory ecosystem has been filling out fast. Our guide to NPU-equipped dev boards covers the silicon side, and purpose-built devices like the reCamera Pro on-device AI camera show where the category is heading. A £9 book that gets a beginner from a bare Pi to a working local transcription pipeline is a useful complement to all of it — the hardware has been ready for a while, and documentation has been the actual bottleneck.

Print copies are available from the Raspberry Pi Press store, Amazon US and UK, and other booksellers, with PDF and ePUB editions alongside. More mini computer coverage here.

Sources: Raspberry Pi — July 21, 2026; Raspberry Pi Press store and bookseller listings — July 21, 2026.