TheLab

• 

Inspiration

This project was inspired by interviews with around 30 professionals from NGOs and emergency crisis organizations. Across humanitarian response, climate adaptation, and disaster recovery, a recurring challenge emerged: decisions about resilience projects are often made too late, using fragmented or outdated information, while realities on the ground evolve in real time.

What it does

ResilienceOS reports, prioritizes, and explains resilience projects in the aftermath of natural catastrophes. By continuously analyzing live data streams, it helps organizations identify which resilience actions should be prioritized now, rather than weeks or months later, turning disaster response into long-term resilience planning.

How we built it

We built ResilienceOS as a real-time, agentic AI system:

Confluent Cloud ingests streaming data, including social media signals and event-driven updates.

Streaming data is enriched and aggregated in real time.

An LLM-powered agent reasons over this data, retrieves contextual knowledge via RAG, and dynamically decides which tools to call.

Social media signals are incorporated as situational awareness to capture emerging community needs and early warnings.

The system outputs ranked, explainable resilience project recommendations.

Challenges we ran into

Handling noisy and unstructured social signals while avoiding misinformation.

Designing agent behavior that is adaptive, not hard-coded.

Balancing real-time responsiveness with explainability and trust.

Integrating multiple tools and data sources while keeping the system scalable and cloud-native.

Accomplishments that we're proud of

Building a fully agentic AI system that reacts to data in motion.

Demonstrating how real-time social and environmental signals can meaningfully influence resilience planning.

Creating an explainable system suitable for NGOs and public-sector decision-makers.

Delivering a working application with a clear path to real-world deployment.

What we learned

Real-time data fundamentally changes how decisions are made — batch analysis is not enough.

Agentic AI is most powerful when tools are selected dynamically, based on context.

Social signals, when used responsibly, provide critical early insights during crises.

Trust and transparency are essential for AI systems in high-stakes environments.

What's next for TheLab

We plan to finalize all proof-of-concepts and deploy the platform with partner organizations, including NGOs and public institutions, to support real-world resilience planning and climate adaptation efforts.

Built With

• cloudrun
• confluent
• fastapi