SentraAI

Real-time disaster relief. Voice in, help out.

Comment

🌍 Inspiration

When disaster strikes, time is critical, but many victims face language barriers, poor connectivity, and a lack of coordination between emergency services. We wanted to build a system that could bridge these gaps using voice, AI, and real-time reasoning. Inspired by the growing power of autonomous agents and large language models, we built SentraAI a voice-activated disaster response system designed to route help faster, smarter, and more inclusively.

🤖 What it does

SentraAI takes a spoken emergency report in any language and instantly:

  • Transcribes the voice input (via ElevenLabs)
  • Translates it to English if needed (via DeepL)
  • Uses Perplexity Sonar Pro to determine the correct response agent (Fire, Police, Medical, NGO)
  • Dispatches structured data to the right agent handler
  • Retrieves contextual info (like nearby hospitals via Apify)
  • Displays the incident on a live map and shows the agent's response

🛠️ How we built it

  • Voice Input: ElevenLabs API for high-accuracy transcription
  • Translation: DeepL API for multilingual input handling
  • AI Reasoning: Perplexity Sonar Pro to determine the appropriate agent
  • Routing Engine: A central dispatcher that formats data into MCP and routes it to modular agent handlers
  • Location Intelligence: Apify API to fetch real-world context (e.g., nearest emergency facilities)
  • Frontend: React + Tailwind CSS with Google Maps integration

⚠️ Challenges we ran into

  • Building a modular agent system that could interact via MCP format
  • Extracting structured location data from unstructured voice input
  • Orchestrating multiple asynchronous APIs in real time
  • Creating a usable voice interface with minimal friction

🏆 Accomplishments that we're proud of

  • Built a fully working multi-agent disaster response system in <6 hours
  • Integrated 3 sponsor tools seamlessly ( DeepL, Perplexity, Apify)
  • Enabled voice-based emergency reporting with real map visualization
  • Designed a dispatcher architecture that mimics real-world emergency communication systems

📚 What we learned

  • How to design and implement agent-to-agent AI workflows using MCP
  • Best practices for real-time multimodal input processing
  • How to combine AI reasoning, translation, voice, and external APIs into a production-ready pipeline
  • The importance of clear, structured prompts when working with reasoning LLMs

🚀 What's next for SentraAI

  • Deploying the system on mobile for use in low-connectivity environments
  • Expanding the agent network (e.g., drones, logistics, shelters)
  • Enhancing agent-to-agent memory and conversation history
  • Partnering with emergency response organizations to pilot the platform in vulnerable regions

Built With

Share this project:

Updates