Inspiration
We were thinking of what if there is a system that can automate a high volume of intake. Help people who got involved in an accident claim insurance faster with the best quality of care.
What it does
SafeGuard AI is an empathetic AI voice agent. When a user calls, the system answers instantly, listens to the caller, and guides them through a structured conversation to collect the most important claim details, including location, incident type, injuries, and damage level. At the same time, it highlights urgent cases for human escalation, logs the claim into connected systems, and stores the full transcript for later review. This makes the process faster, more supportive, and more efficient for both the customer and the insurer.
How we built it
We built SafeGuard AI as a real-time voice AI pipeline. Twilio handles inbound phone calls, Groq Whisper converts speech into text, and DeepSeek manages the conversation by balancing empathy with structured data collection. ElevenLabs turns the AI response back into natural voice. On the backend, we used FastAPI to orchestrate the flow and SQLite to store transcripts and records. We also connected Google Sheets for automatic claim logging and built a Streamlit dashboard for live monitoring and emergency visibility.
Challenges we ran into
We had to think carefully about latency, because even a short delay can make a voice conversation feel unnatural. Another challenge was designing the logic for escalation so that urgent cases, especially those involving injuries, could be identified and surfaced immediately.
What we learned
We learned that building voice AI for high-stress situations is very different from building a normal conversational assistant. Empathy, timing, and trust matter just as much as technical accuracy.
What's next for Safeguard AI
The next step is to make SafeGuard AI even more production-ready. We want to improve claim classification accuracy, strengthen escalation logic, and integrate with enterprise insurance claim platforms beyond spreadsheets and local storage. We also want to add multilingual support, better analytics for adjusters, and stronger compliance and security features. In the long term, we see this architecture expanding beyond insurance into other high-stress intake scenarios like roadside assistance, healthcare triage, and emergency support.
Log in or sign up for Devpost to join the conversation.