Clarity

Inspiration

Communication is more than just words — it includes tone, intent, and emotion. In fast-paced meetings and virtual conversations, people often miss critical meaning. This problem is even more significant for individuals with autism, ADHD, or language-processing challenges. We were inspired to build Clarity to bridge the gap between and , helping people better understand conversations in real time.

What it does

Clarity is a real-time meeting assistant that listens to conversations and provides:

• Simplified summaries of what was just said
  
• Emotional context (e.g., urgency, frustration, calmness)
  
• Automatic extraction of action items and deadlines
  
• Suggested calendar updates based on tasks
  

How we built it

AI pipeline (index.js)

• The pipeline starts with an inputProcessor that prepares the incoming transcript and emotion data.
• From there, we run multiple tasks in parallel using Promise.all() for speed:
  • simplifier for summary generation
  • explainer for vocabulary/context support
  • emotionDetector for classifying across 10 emotions
  • replyGenerator for generating 1–3 suggested replies
  • taskExtractor for identifying action items
• After that, a responseGenerator creates a supportive final response.

• The results are then stored using store.save() into a memoryStore / fileStore.

  • LLM integration

• The processing is powered by Gemini 2.5 Flash through the Generative Language API, which enables fast real-time inference.

Overall flow:
Camera + Microphone → Browser Detection/Transcription → Express Backend → Parallel AI Pipeline → Overlay Results + Memory Storage

Challenges we ran into

• Maintaining low latency for real-time feedback
  
• Handling noisy or overlapping speech input
  
• Accurately detecting emotion from limited context
  
• Extracting structured action items from unstructured conversations
  
• Designing a UI that provides value without overwhelming the user
  

Accomplishments that we're proud of

• Built a working real-time audio → insight pipeline
  
• Successfully simplified complex speech into concise summaries
  
• Implemented emotion detection in an accessible and intuitive way
  
• Automated action item tracking from live conversations
  
• Designed an accessibility-first solution that benefits a wide audience

What we learned

• Real-time systems require careful trade-offs between speed and accuracy
  
• Simplification is more challenging than basic summarization
  
• Emotion detection is nuanced and context-dependent
  
• Accessibility-focused design improves usability for everyone
  
• Clear UX is critical when presenting complex AI outputs
  

What's next for Clarity

• Improve emotion detection using multimodal inputs (voice + facial cues)
  
• Integrate with platforms like Zoom, Microsoft Teams, and Google Meet
  
• Add speaker identification and role-based task assignment
  
• Personalize summaries based on user preferences
  
• Optimize for large-scale deployment and enterprise use
  

Built With

• css
• html
• javascript