Hands-on: real-time face tracking in the browser, no build tools
Build a working 478-landmark face tracker in a single HTML file: MediaPipe from a CDN, getUserMedia, and a canvas overlay. No build tools, no server.
Signals
Tutorials
Working code or it didn’t happen. Step-by-step builds of real vision and AI systems. Each tutorial is self-contained and ends with something running.
01
In this signal
Object detection in the browser with Transformers.js (no server)
Run object detection entirely in the browser with Hugging Face Transformers.js and WebGPU: no server, no build tools. One file, your webcam, live boxes.
Real-time hand & pose tracking in the browser (a MediaPipe follow-up)
Track 21 hand landmarks and 33 pose landmarks in the browser with MediaPipe Tasks: one HTML file, a webcam, and a pinch-to-click gesture. No build tools.
Reasoning models, demystified: build a test-time-compute loop
Test-time compute, explained: what DeepSeek-R1 and o-series models actually do, then build a self-consistency loop that makes any model reason harder.
Fine-tune without a GPU farm: LoRA and QLoRA, explained with code
Fine-tune a modern LLM on a single GPU. LoRA and QLoRA from first principles, with runnable code and a clear-eyed guide to when fine-tuning is the wrong tool.
Reliable JSON from an LLM: structured outputs and function calling
Stop regex-parsing model output. Use JSON Schema structured outputs and function calling for reliable, typed data from an LLM, with validation and retries.
Give your model tools: your first Model Context Protocol (MCP) server
MCP is the open standard for giving LLMs tools and context. Build a minimal MCP server, connect a client, expose your first tool. Security caveats included.
Retrieval-augmented generation from scratch, in one file
Build a working RAG system in one Python file: embeddings, cosine search, and a grounded prompt. No framework. Own it before you reach for LangChain.
Run a capable LLM on your laptop: a no-nonsense local setup
No API key, no GPU rental. Run a capable open LLM on your laptop in ten minutes, and learn the quantization setting that makes it feel instant.