OpenMic

OpenMic is the always-on voice feedback system that captures real-time insights when users want to share them. No more long surveys—just speak, and AI turns it into actionable data instantly.

Comment

OpenMic: The Always-On Voice Feedback System

Try it out...

You are a product manager at Eleven Labs that created a voice agent to collect user feedback to learn more about Devs' experience with Eleven Labs' products... more specifically around realism, latency of the voice... *Here's our mock session: voice agent demo *

Team

  • Rayane Achich – Frontend development, LLM prompt engineering
  • Neil Achich – Backend development, voice AI and Supabase integrations

Inspiration

Collecting feedback is broken. Traditional surveys are slow, have low participation rates, and fail to capture meaningful insights. Teams struggle with:

  • Inefficient processes – Surveys take weeks, focus groups are costly.
  • Low engagement – Response rates are often below 1%.
  • Shallow insights – Written responses lack the nuance of real conversations.

We wanted a system that makes feedback fast, natural, and structured—allowing every voice to be heard without friction. That’s why we built OpenMic, an AI-powered voice agent that captures, analyzes, and organizes feedback in real time.

What It Does

OpenMic provides a seamless way for teams, brands, and researchers to collect user insights without traditional surveys:

  1. Always-On Voice Hotline – Users can share feedback anytime.
  2. AI-Powered Summarization – Conversations are transcribed and structured into key themes.
  3. Live & Asynchronous Options – Works for real-time brainstorming or post-event reflections.
  4. Higher Engagement – Speaking is easier than typing, leading to richer insights.

By reducing friction, OpenMic delivers faster, deeper, and more actionable feedback than traditional methods.

How We Built It

OpenMic is powered by a seamless integration of AI, voice processing, and a scalable web platform to deliver instant, structured feedback. Here's how it works:

  1. Survey Creation:

    • A logged-in user creates a survey prompt.
    • OpenAI's LLM helps generate structured questions.

  2. Voice Feedback Collection:

    • An unlogged customer starts a feedback session.
    • ElevenLabs converts the survey prompt into natural-sounding speech (TTS).
    • The customer responds in real-time, and Deepgram transcribes the audio (STT).

  3. AI-Powered Summarization:

    • Responses are processed through OpenAI to extract key themes.
    • Feedback is structured into meaningful insights.

  4. Survey Results Delivery:

    • Results are displayed as structured text for easy analysis.
    • Optionally, ElevenLabs can convert results into voice for audio-based insights.

Tech Stack:

  • Frontend: Next.js for a smooth user experience.
  • Backend: Supabase for data management, Vercel for hosting.
  • Voice AI: ElevenLabs (TTS) and Deepgram (STT) for real-time voice interactions.
  • AI Processing: OpenAI's LLM for structured feedback analysis.
  • Developer Tools: Cursor for coding, Lovable for UI enhancements.
  • GitHub Repo: OpenMic Source Code

By combining real-time voice AI with structured insights, OpenMic transforms the way teams collect and analyze feedback—faster, easier, and more engaging than traditional surveys.

Challenges We Faced

  1. Ensuring consistent JSON formatting for surveys – Structuring survey content into a forced JSON format so that the voice agent understands the context every time proved challenging. We had to refine prompt engineering to make it work reliably.

  2. Accurate JSON conversion for transcriptions – When converting voice responses back into structured JSON, maintaining accuracy and extracting key themes required fine-tuning our AI processing pipeline.

  3. Making the voice agent engaging – Designing a voice agent that keeps users engaged while ensuring survey objectives are met required careful scripting and iteration on tone, pacing, and interaction flow.

What We Learned

  • Voice feedback is more expressive – Users communicate more nuance through speech than text.
  • Summarization is key – A transcript alone isn’t useful; structured insights drive action.
  • Frictionless experiences boost engagement – Making feedback effortless increases participation.

What’s Next?

  • Will build the capability to voice chat with survey results...
  • Automating feedback-driven action items for product teams

Built With

  • cursor
  • deepgram
  • elevenlabs
  • lovable
  • nextjs
  • openai
  • supabase
  • vercel
Share this project:

Updates