Inspiration
Customer support AI often focuses on resolution, not emotion. We noticed that most voice bots continue conversations even when users are frustrated, leading to silent churn. The idea was to build a voice AI that doesn’t just respond, but knows when to step back and bring in human help.
What it does
Mobile Assist is an emotion-aware voice AI that listens beyond words. It detects frustration, dissatisfaction, and confusion in real time during conversations.
If negative sentiment is detected, it gently pauses the interaction and assures the user that a human executive will follow up. Every interaction, resolved or escalated, automatically generates a support ticket with emotional context for operational visibility.
How we built it
We built Mobile Assist using:
- Voice interaction powered by Deepgram
- Real-time sentiment monitoring via Deepgram Audio Intelligence
- Function calling to trigger external actions
- Automated ticket creation using Trello
The system continuously analyzes tone and conversational signals while the AI speaks, enabling parallel sentiment detection and real-time escalation.
Challenges we ran into
- Running sentiment analysis in parallel without interrupting conversations
- Defining thresholds for dissatisfaction detection
- Designing a soft-stop mechanism instead of abrupt termination
- Ensuring tickets captured both issue and emotional state
- Handling complaint scenarios with faster escalation sensitivity
Accomplishments that we're proud of
- Built real-time frustration detection into voice AI
- Enabled graceful AI-to-human handoff
- Ensured 100% interaction tracking via auto-ticketing
- Created emotional context tagging for support teams
- Prevented unresolved negative experiences from going unnoticed
What we learned
- Resolution alone does not equal satisfaction
- Emotional intelligence is critical for conversational AI
- AI needs escalation boundaries
- Operational visibility improves trust in automation
What's next for Mobile Assist
- Smarter escalation prediction
- CRM integrations
- Multilingual sentiment detection
- Analytics dashboard for emotional trends
- Adaptive response behavior based on user mood
Built With
- deepgram
- drizzle
- node.js(express)
- postgresql
- python(fastapi)
- react-native
- trello
Log in or sign up for Devpost to join the conversation.