VISOIC – Voice-First AI Assistant, Realistic Video Responses

VISOIC is a fully working AI assistant that lets users control tools like Gmail, HubSpot, Slack, and Notion using voice or text. It replies with both voice and realistic human-like video from Tavus.

Comment

About the Project

🧠 “What if everyone could control their digital life using just voice?”

VISOIC was inspired by the need for more inclusive technology. I wanted to build an assistant that users with limited mobility or visual impairments could rely on — one that listens, understands, and responds in a way that feels natural.

What It Does

🗣️ “Say 'Hey Visoic' followed by any request — and watch it respond.”

VISOIC is a voice-first AI assistant that interprets natural language commands and executes real-world actions like sending emails, navigating pages, or booking flights. It provides feedback through realistic voice (via ElevenLabs) and human-like video responses (via Tavus.io).

How I Built It

🛠️ “Built entirely with Bolt.new, React, Pica AI, Supabase, ElevenLabs, and Tavus.io.”

I used:

  • Bolt.new – For hackathon-compliant hosting
  • React + TypeScript – For a scalable frontend
  • Pica AI – For agentic command parsing
  • Supabase – For authentication and logging
  • ElevenLabs – For natural-sounding TTS
  • Tavus.io – For realistic avatar video responses

Everything connects seamlessly to create a fluid, accessible experience.

Challenges I Ran Into

⚠️ “Getting Pica AI to understand complex, context-rich commands wasn't easy.”

The biggest challenges were:

  • Making sure wake word detection (“Hey Visoic”) was reliable
  • Ensuring agentic command interpretation was accurate across contexts
  • Syncing audio with Tavus video responses
  • Handling accessibility edge cases for screen readers and keyboard-only navigation

Accomplishments I'm Proud Of

🏆 “A truly hands-free assistant — no mouse, no touch, just voice.”

I’m proud of building a working MVP where:

  • Users can navigate and execute tasks using only voice
  • The assistant responds via realistic audio and video
  • Everything is fully accessible and responsive

This proves that AI can be inclusive, not just functional.

What I Learned

💡 “Voice-first design changes everything — from UX to backend logic.”

I learned how to:

  • Use agentic AI for multi-step task execution
  • Build fully accessible UIs
  • Integrate real-time speech and video APIs
  • Structure modular components for scalability
  • Deploy a production-ready app using Bolt.new

What’s Next for VISOIC

🚀 “From voice assistant to full AI companion — human-like and always ready.”

Next steps include:

  • Adding image input interpretation
  • Supporting human-in-the-loop fallback
  • Implementing blink detection for ultra-low mobility users
  • Enabling multi-user sessions
  • Expanding integrations to calendar, CRM, and productivity tools

VISOIC will evolve into a full digital life assistant — built for everyone.

Built With

  • agent
  • ai
  • authentication
  • avatar
  • backend
  • bolt.new
  • command
  • commandlogs
  • components
  • cvi
  • database
  • deployment
  • design
  • edgefunctions
  • elevenlabs
  • engine
  • file
  • framework
  • hosting
  • language
  • logging
  • lucide
  • management
  • mobilefirst
  • oauth
  • parsing
  • picaai
  • postgresql
  • react
  • reactrouter
  • recognition
  • response
  • responsive
  • routing
  • state
  • storage
  • styling
  • supabase
  • tailwind
  • tavus
  • tts
  • typescript
  • ui
  • video
  • voice
  • webspeech
  • zustand
Share this project:

Updates