LLM & Agents · 2026
Generative AI Experience Hub for Flagship Stores
How do you make AI tangible for people walking into a store? You let them talk to it, play with it, and take the result home on their phone.
→From Electron prototype to multi-store web app with voice agents & generative portraits.
> visitor: "Can I talk to Einstein?"
agent.voice.connect(persona)
portrait.style("anime") → QR ✓ Highlights
- ✦Prototype on real kiosk hardware within weeks
- ✦Speech-to-speech voice agents, incl. historical persona avatars
- ✦Webcam portraits in a growing library of AI styles
- ✦Live, web-grounded answers to visitors' questions
- ✦Rolled out per store with QR hand-off to visitors' phones
The challenge
A major German telecommunications brand wanted to show what AI can do. Not on a slide. In their flagship stores.
For real people. All day. Without a developer standing next to it.
No pressure. 😅
What I built
Step 1: Prototype fast. An Electron app on real kiosk hardware, within weeks. Test with real visitors, learn what works.
Step 2: Build it properly. Rebuilt as a React 19 + Vite web app with a small Node server and Docker deployment. Configurable per store.
What visitors can do:
- 🎙️ Talk to AI. Speech-to-speech voice agents (ElevenLabs), including persona avatars you can interview
- 📸 Get an AI portrait. Webcam capture, then a catalogue of AI styles via image APIs like Picsart and Replicate
- 🔎 Ask anything. Live, web-grounded answers via the Perplexity API
- 📱 Take it home. Results hand off to your phone via QR code
The lesson
Multiple AI providers. Flaky store Wi-Fi. Non-technical users. A brand that can’t afford an awkward demo.
The fix: every AI capability sits behind a small interface. When a provider deprecates a model, it’s a config change, not a crisis.
AI in the real world isn’t about the model. It’s about everything around it.
