SassyAI: The Sarcastic AI Assistant

#devchallenge #awschallenge #ai #webdev

What I Built
SassyAI: The Judgy Character Assistant

SassyAI started as an interactive, sarcastic CLI assistant built for developers and tech enthusiasts who wanted a humorous break from serious tools. It didn’t just answer
questions, it judged them.

Since then, the project has evolved into SassyAI V2, a web-based character assistant powered by an LLM, with a visual personality, attitude-driven responses, and a more polished
user experience.

SassyAI is still rude, sarcastic, impatient, and weirdly useful.
It just has better lighting now.

Update: V2 is live 🚀

A new version is now available as a web character assistant powered by an LLM, with:

a FastAPI backend
a single-page frontend
a character-driven UI
attitude-based visual states
a versioned sarcastic persona
safety handling and fallback behavior

Repo (V2): https://github.com/hericlibong/SassyAI
Live demo (Hugging Face): https://ricolajazz-sassy-ai.hf.space

The original CLI version is still available here:

Repo (V1): https://github.com/hericlibong/SassyAI-v1

Features

V2 Web Experience

Character-driven interface: SassyAI is no longer just a message bubble. It appears as a visual character that reacts to the conversation.
Attitude-based visuals: The character can switch between states like idle, thinking, sarcastic, serious, refused, and fallback.
LLM-powered responses: The assistant now uses a real model provider through a backend abstraction layer.
Versioned persona: The sarcastic behavior is controlled through dedicated persona files, making tone changes traceable and maintainable.
Tone control: Users can choose between Soft, Spicy, and Brutal but useful.
Typing/reveal effect: Assistant replies appear progressively, with a skip option.
Quick prompt chips: Users can start common interactions quickly.
History panel: The latest exchange stays central, while older messages remain available in a secondary panel.
Safety and fallback handling: Unsafe requests are refused or neutralized, and provider failures return a controlled fallback response.
Hugging Face deployment: The app is deployed as a Docker Space with FastAPI serving both the frontend and backend.