Stanect — AI Voice Companion Platform
AI voice companion I built end-to-end so people can practice conversations and build speaking confidence — from landing and onboarding through live sessions and subscriptions.
AI voice companion I built end-to-end so people can practice conversations and build speaking confidence — from landing and onboarding through live sessions and subscriptions.
Role:
Founder & Full-Stack Engineer
Ownership:
Built the entire product end-to-end
Users:
200+ users
Latency:
Sub-500ms real-time voice interactions
Performance:
About 20% Lighthouse performance improvement
Core Stack:
Focus:
Real-time voice, reliable onboarding, scalable architecture
Many people want to improve their communication skills but lack safe environments to practice conversations. I explored whether an AI voice companion could create a low-pressure space where users can experiment, practice responses, and build confidence speaking to others.
Companion creation flow
Preview-driven configuration
Real-time conversation sessions (WebSockets, sub-500ms)
Profile management
Subscription model for premium access




Persistent onboarding state so users do not lose progress
WebSockets and streaming for sub-500ms voice interactions
React Server Components and a modular architecture for maintainability
Strict enough validation to keep bad data out of onboarding without blocking the flow
Frontend, data, backend, and voice are split so each layer can change without rewriting the rest.
Frontend:
Next.js
React
TypeScript
Tailwind CSS
Framer Motion
State and data management:
Zustand
TanStack Query
Zod
Backend infrastructure:
Node.js
Supabase
Clerk
WebSockets
Voice and AI processing:
Vapi for voice processing
OpenAI for conversation generation
WebSockets sit on the backend layer so voice stays under 500ms.