Let’s build K.I.T.T. with JavaScript
summary
Builds a browser K.I.T.T. using JavaScript that actually runs offline and claims privacy by default, combining transforms.js + Whisper + kakoro for speech, Gemma 2 LLM for reasoning, WebAssembly/WebGPU for performance and WebBluetooth to drive a Lego Arduino Nano RP2040 car; stores memories in IndexedDB with vector search from miniLM6v2 embeddings and a RAG + function‑calling agent. Processed 6,275 scenes, 3.1M input tokens → 303k output tokens; the car responds and higher‑level decisions work though small models sometimes request duplicate function calls.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 2:31What is K.I.T.T.
- 6:24Listening and talking
- 9:54LLMs in browser
- 11:58K.I.T.T.'s personality
- 13:20K.I.T.T.'s memories
- 20:26Agentic RAG system
- 23:57K.I.T.T. with Lego
- 27:51Project summary
- 28:11Why build K.I.T.T.
related talks
43mLet’s build K.I.T.T. with JavaScript
40mLet’s build K I T T with JavaScript
43mLet’s build K.I.T.T. with JavaScript
33mLet’s Build K.I.T.T. with JavaScript
30mLook ma, no hands! Multimodal AI-Agents in the browser
37mJavaScript On Microcontrollers
30mUnleashing AI in Your Web Apps
22mBoring but Effective: WebAI That Ships