Building a Candy AI-Inspired Chatbot Using Gemini AI

Gemini API
,
1

Hey everyone

I’m currently working with NSFW Coder on an educational AI development project inspired by systems like Candy AI.
Our goal is to explore how AI companions simulate personality, emotional tone, and contextual memory — and how these techniques can be adapted for ethical, educational, and interactive chatbot applications.

We’re using Gemini AI as the main conversational model and focusing on:

  • Developing realistic, personality-driven responses

  • Implementing memory persistence using vector databases

  • Reducing response latency in real-time chat systems

Here’s our tech stack so far:

  • Frontend: React + Tailwind

  • Backend: Node.js + Express

  • AI Engine: Gemini API

  • Memory System: ChromaDB

  • Hosting: Firebase / Vercel

Our current challenges include:

  1. Response delays: Gemini sometimes takes up to 10–15 seconds for replies.

  2. Memory resets: The local memory system (Chroma) isn’t maintaining embeddings between sessions.

  3. Consistency in personality: Emotional responses vary from robotic to overly expressive.

We’re using this project as a learning experience to understand practical AI architecture and ethical design in conversational systems.

If anyone has experience with AI companion frameworks, context memory optimization, or emotion modeling in chatbots, your insights would be super helpful!