• 

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for Smart Disaster Response Agent

🌍 Project Story – Smart Disaster Response Agent 📌 About the Project

The Smart Disaster Response Agent is an AI-powered system designed to help local authorities and communities respond faster during natural disasters such as floods, earthquakes, and cyclones. By combining real-time data ingestion, vector + full-text search on TiDB Serverless, LLM-powered analysis, and external tool integration (maps, weather, and alert systems), the agent transforms raw information into life-saving, actionable insights.

💡 Inspiration

Natural disasters often cause chaos, where critical information is scattered across social media, sensor feeds, and news outlets. During emergencies, delays in gathering and analyzing this data can cost lives. We were inspired by the idea:

“What if AI could act as a real-time assistant for disaster response teams, helping them cut through noise and act quickly?”

This led us to build a multi-step agent that not only searches and summarizes data but also takes meaningful action by triggering alerts and mapping affected areas.

🛠️ How We Built It

Our architecture chains multiple building blocks into a single workflow:

Ingest & Index Data

Collected sample datasets: mock flood alerts, social media posts, IoT sensor readings.

Ingested into TiDB Serverless, storing both structured logs and unstructured text.

Created vector embeddings for similarity search and full-text indexes for keyword queries.

Search & Retrieval

Implemented hybrid search:

Relevance Score

𝛼 ⋅ VectorSim + ( 1 − 𝛼 ) ⋅ KeywordMatch Relevance Score=α⋅VectorSim+(1−α)⋅KeywordMatch

This allowed both semantic similarity and keyword precision.

LLM Analysis

Used an LLM to cluster similar alerts, summarize reports, and highlight key actions.

Example: “15 alerts indicate flooding near Singanallur; traffic blocked on Road X.”

External Tools

Integrated Mapping APIs to visualize hotspots.

Connected to Slack/WhatsApp Bot APIs to auto-notify response teams.

Pulled data from a Weather API to enhance decision-making.

🚧 Challenges We Faced

Data Diversity: Handling mixed inputs (structured sensor data + noisy text data).

Realism vs. Demo Simplicity: Balancing between a practical workflow and a hackathon-friendly prototype.

Workflow Chaining: Ensuring smooth transitions between vector search, LLM calls, and external APIs.

Latency: Optimizing query response time to be fast enough for emergency contexts.

📚 What We Learned

How to effectively combine vector search + full-text search for robust retrieval.

Designing agentic multi-step workflows that go beyond RAG demos.

The power of TiDB Serverless in managing diverse, real-time datasets.

Importance of human-centered design when building tools for disaster management.

✨ With this project, we learned that AI agents are not just about answering questions—they can save time, coordinate responses, and ultimately save lives.

Built With

• cloud
• clustering
• data-storage-tidb-cloud-?-scalable-backend-database-apis-&-tools:-openai-/-llm-apis-?-summarization
• docker
• fastapi
• google-maps
• javascript
• javascript-(for-frontend)-frameworks:-fastapi-?-backend-agent-workflow-orchestration-streamlit-?-demo-dashboard-&-visualization-databases-/-cloud-services:-tidb-serverless-?-vector-+-full-text-search
• openai
• openweather
• python
• slack
• streamlit
• tidb
• whatsapp