Let’s build K.I.T.T. with JavaScript
summary
Recreates K.I.T.T. entirely in-browser: transformers.js for on-device LLMs, WebGPU/WebAssembly for speed, Whisper and kakoro for speech I/O, and WebBluetooth to drive a Lego car backed by an Arduino Nano RP2040. Uses IndexedDB, miniLM L6 v2 embeddings and a RAG pipeline with function-calling (XML) to query 6,275 scenes—processing about 3.1 million input tokens and generating 303,000 output tokens. Runs locally at ~70 tokens/s ingest and ~31 tokens/s generate; the demo works though hardware (battery) and small-model quirks require extra app-side checks.
AI-written summary · about
chapters (11)
- 0:00Building K.I.T.T. in browser
- 3:15GPU for ML in JavaScript
- 8:32Speech with Whisper & Kakoro
- 12:10LLMs on device with WebLM
- 14:28K.I.T.T.'s personality via prompts
- 17:37Storing Night Rider memories
- 25:07Function calling for memory recall
- 29:43Lego K.I.T.T. via Web Bluetooth
- 33:15Movement and decisions for car
- 36:48Why build K.I.T.T. & on-device AI
- 39:18Q&A
related talks
43mLet’s build K.I.T.T. with JavaScript
30mLet’s build K.I.T.T. with JavaScript
33mLet’s Build K.I.T.T. with JavaScript
40mLet’s build K I T T with JavaScript
30mLook ma, no hands! Multimodal AI-Agents in the browser
22mBoring but Effective: WebAI That Ships
25mTransformers.js: Building Next-Generation WebAI Applications
20mFrom ML to LLM: On-Device AI in the Browser