About the Project CrisisLens AI Inspiration
During floods, road damage, garbage overflow, water leakage, and other local emergencies, people often report problems through different channels. However, these reports are usually unstructured, duplicated, and difficult to prioritize.
We were inspired by a simple question:
What if AI could turn thousands of scattered citizen reports into a clear list of incidents that actually need immediate action?
This led us to build CrisisLens AI, an AI-powered platform that converts citizen-generated text, images, and voice reports into structured, prioritized incidents.
What We Built
CrisisLens AI allows users to submit a photo, text description, or voice report of a local problem.
Our system then:
Extracts and understands the report using AI. Classifies the type of incident. Analyzes the severity and potential impact. Identifies duplicate or related reports. Groups nearby reports into a single incident. Prioritizes incidents based on urgency. Generates an AI-powered action recommendation. Displays the incidents through a centralized dashboard.
For example, if 20 people report flooding on the same road, the system can recognize them as a single high-priority incident instead of treating them as 20 independent complaints.
How We Built It
The system follows a multi-stage architecture:
Citizen Report ↓ Text / Image / Voice ↓ AI Analysis ↓ Incident Classification ↓ Severity Assessment ↓ Duplicate Detection ↓ Location-Based Clustering ↓ Incident Prioritization ↓ AI Action Recommendation ↓ Response Dashboard
We use Python-based AI processing for data analysis and machine-learning tasks, an LLM/vision model for understanding unstructured reports, PostgreSQL for structured incident data, and a backend API to connect the AI pipeline with the dashboard.
The system is designed to preserve the original reports while generating structured information such as:
Incident Type: Urban Flooding Severity: Critical Confidence: 94% Related Reports: 17 Priority: Immediate Recommended Action: Deploy drainage and emergency response team What We Learned
Building CrisisLens AI helped us understand that solving a real-world problem with AI is not just about using an LLM.
We learned how to combine:
Multimodal AI Natural language processing Data cleaning and validation Semantic similarity Location-based clustering Machine-learning-based prioritization Database design REST APIs AI-generated recommendations
Most importantly, we learned how to transform unstructured, noisy real-world data into structured information that can support decision-making.
Challenges We Faced
One of our biggest challenges was duplicate detection. Two people may describe the same incident using completely different words. Simple keyword matching would treat them as separate incidents, so we explored semantic similarity combined with geographical proximity.
Another challenge was severity estimation. A report saying "water on the road" does not necessarily indicate the same level of danger as "water has entered multiple houses." We therefore considered multiple signals rather than relying only on the text.
We also had to design the system so that AI-generated recommendations are explainable and actionable, rather than simply producing generic chatbot responses.
Impact
CrisisLens AI aims to reduce the gap between reporting a problem and understanding its real-world urgency.
Instead of overwhelming authorities with hundreds of disconnected complaints, the platform can provide a consolidated view:
What happened? Where did it happen? How serious is it? How many people are affected? What should be prioritized?
Our goal is to demonstrate how AI can move beyond conversation and become a practical real-world decision-support system for safer and more responsive communities.
Built With
- ai-agents
- cloud
- computer-vision
- data-processing
- embeddings
- fastapi
- generative-ai
- geospatial-analysis
- github
- large-language-models-(llm)
- machine-learning
- natural-language-processing-(nlp)
- pandas
- postgresql
- python
- react
- rest-api
- semantic-search
- spring-boot
Log in or sign up for Devpost to join the conversation.