Let’s Build K.I.T.T. With JavaScript

Nico is a frontend developer, part of the Google Developer Experts program in web technologies and machine learning and AI, who presented at JS Nation about recreating the iconic car KIT from Knight Rider using JavaScript.

KIT is short for Knight Industries 2000, an AI-powered car from the 1980s TV series Knight Rider, recreated by Nico in JavaScript as part of a side project.

The key features of KIT include natural language communication, intelligence, a personality, a history with Michael Knight, and a physical form, all of which Nico aimed to recreate using JavaScript and machine learning.

Nico uses Transformers.js for processing language, Whisper for transcribing spoken words, Kokoro.js for text-to-speech conversion, and large language models (LLMs) for intelligence and conversation, including prompt engineering for personality.

Nico uses Transformers.js and the LLM library to implement machine learning models directly in the browser, allowing KIT to process language and interact intelligently.

Vector databases are used to store and retrieve scenes from the Knight Rider series based on semantic similarity, allowing KIT to reference past events and provide contextually relevant responses.

Tool calling is implemented by allowing the application to execute functions based on structured output generated by the LLM, enabling KIT to perform actions like searching for specific scenes.

Nico uses a Lego model of KIT with an Arduino microcontroller and Bluetooth low energy communication to control the car's movements via the browser.

Nico undertook the project for fun and to explore the capabilities of JavaScript and machine learning, enjoying the process of learning and building new things.

Nico faced challenges such as ensuring function calling consistency with smaller models, parsing screenplay data for vector databases, and managing performance constraints with available hardware.