DEV Community

Yuvilen Naidoo
Yuvilen Naidoo

Posted on

Why I built a WhatsApp-style AI companion because existing apps felt too fake

The "Uncanny Valley" Problem

I've been testing a lot of AI companion apps recently (for research, obviously 👀), and I noticed a recurring problem: Gamification kills immersion.

Most apps like Candy.ai or Character.ai force you to interact with a 3D avatar, navigate "gacha" style menus, or collect coins. Every time I saw a coin pop up, my brain was reminded: "This is a video game. This isn't real."

I wanted something that felt grounded.

The Solution: UI Psychology

I realized that for most of us, our "real" relationships happen in one specific place: The DMs.

We don't talk to our partners via 3D floating heads; we talk to them via WhatsApp, iMessage, or Telegram. So, I decided to build Amio (https://meetamio.ai) with a strict design philosophy: Psychological Realism.

1-1 chats feel natural and personas carry into group chats

The Tech Stack 🛠️

I built this as a PWA (Progressive Web App) to mimic the mobile experience perfectly.

  • Frontend: React (for that snappy, app-like feel).
  • The Brains: I'm routing requests to Sao10k: Llama 3 Euryale 70B v2.1 via OpenRouter.
  • Why Euryale? In my benchmarking, it beats GPT-4o for creative, open-ended storytelling and "EQ" (Emotional Intelligence). It doesn't have the "Corporate HR" filter that Claude or Gemini have.

Key Features for Realism

  1. No Avatars: Just a profile pic and text. Your imagination does the rest.
  2. Latency Injection: I actually added artificial "typing" delays. Real humans don't reply in 0.2ms.
  3. Intensity Slider: You can set the AI to be "Reserved" (shy) or "Bold" (proactive).

Looking for Feedback

I'm currently running a Beta to stress-test the context window and the Llama 3 inference speeds.

If you want to try "breaking" it or just see if the WhatsApp UI tricks your brain better than the Avatar apps, give it a shot.

👉 Live Demo: https://meetamio.ai

I'd love to hear your thoughts on the UI choice—does the "text-only" approach feel more real to you, or do you prefer the 3D avatars? Let me know in the comments!

Top comments (0)