Building an Agentic AI Framework for Disaster Response

Our team participated in the global IBM TechXchange Hackathon, focused on developing agentic AI solutions using IBM’s watsonX platform to address real-world challenges. The objective was clear: not to build another chatbot, but to architect an AI system capable of coordinating multiple agents, reasoning over context, and responding dynamically in complex environments.

Out of thousands of global submissions, our project ranked among the Top 50 highest-scoring qualified entries worldwide. As recognition, our team received full-access tickets to the IBM TechXchange Conference, held in Orlando, Florida. Roni Noueihed and Hunter Price (Senior Project Engineers) represented the Ingenics team onsite, engaging with IBM’s latest advancements in enterprise AI, agent orchestration, and applied digital transformation. This achievement reflects something central to how we work: designing AI solutions that are innovative, yet grounded in operational reality.

• The IBM TechXchange Hackathon created an opportunity to evaluate how agentic AI patterns can support high-impact scenarios using IBM watsonX. Rather than optimizing for novelty alone, the team intentionally selected a use case that mirrors real operational complexity and urgency: natural disaster response.
• Flooding, hurricanes, and wildfires create rapidly evolving conditions, fragmented information sources, and significant cognitive load for affected individuals. These characteristics make disaster response a strong test environment for orchestrated AI systems designed to support timely, informed decision-making.
• This focus allowed the team to prioritize practical system design — how agents reason, how they coordinate, and how they remain grounded in authoritative data when accuracy and trust are critical.

For the hackathon, the team focused on a high-impact, time-critical use case: enabling faster and more informed responses to natural disasters.

The solution was intentionally designed as citizen-facing, prioritizing clarity, reliability, and actionable guidance for individuals navigating emergency situations.

Rather than relying on a single AI agent, the architecture implemented a coordinated multi-agent framework led by a master orchestration agent — reflecting emerging best practices in agentic system design.

At the core of the solution is a master agent responsible for:

• Interpreting user input and contextual cues
• Identifying the type of emergency being described
• Dynamically routing requests to the appropriate specialized disaster agent
• Consolidating responses into a single, prioritized set of recommendations

The orchestration logic was prompt-driven, allowing the master agent to coordinate behavior through structured reasoning instructions rather than rigid rules. This approach enabled flexibility while maintaining transparency and control — a critical consideration for citizen-facing AI systems.

Each specialized agent (flooding, hurricanes, wildfires) was optimized with domain-specific prompts, enabling focused analysis of risks, response actions, and safety guidance relevant to each scenario.

To support accuracy and trustworthiness, the agents were built on a Retrieval-Augmented Generation (RAG) architecture using IBM watsonX document and vector storage capabilities.

Authoritative public information was collected, structured, and indexed from trusted sources such as:

• FEMA
• National Hurricane Center
• CDC
• American Red Cross

This allowed the agents to retrieve region-specific and event-specific guidance — including evacuation recommendations, shelter options, health precautions, and post-event actions — before generating responses.

By grounding outputs in curated external knowledge, the system avoided generic recommendations and instead delivered context-aware, situationally relevant guidance.

In addition to structured guidance, the solution incorporated near-real-time situational awareness through publicly available, standards-based data interfaces that provide updates on evolving emergency conditions.

This enabled the agents to:

• Account for the latest developments in active disasters
• Adjust guidance as conditions change
• Provide earlier, more informed recommendations to support faster and safer action

The combination of live data signals with retrieval-based knowledge grounding demonstrated how agentic AI can enhance decision-making in rapidly changing, high-risk environments.

While the solution did not implement long-term memory across users, it maintained short-lived session context within each interaction. This allowed the agent to:

• Preserve key details shared during a conversation
• Avoid redundant questions
• Maintain continuity as situations were clarified

This lightweight session awareness balances usability with architectural simplicity — particularly important for emergency tools where speed and clarity outweigh deep personalization.

This project highlights our ability to:

• Design and implement orchestrated, multi-agent AI systems
• Leverage IBM watsonX foundation models in applied environments
• Build prompt-driven orchestration logic that remains explainable and controlled
• Apply RAG architectures to ground AI responses in authoritative data
• Integrate real-time external data sources into AI workflows
• Design solutions aligned with real-world constraints and user trust

Most importantly, it reflects how we approach AI development: structured, pragmatic, and focused on delivering measurable value rather than experimentation alone.

Achieving this outcome required close collaboration across the team, combining technical experimentation, architectural design, and rapid iteration within the constraints of the hackathon environment. Team members worked together to define agent responsibilities, refine orchestration prompts, curate trusted data sources, and align on design decisions that balanced flexibility with reliability.

The solution evolved through shared problem-solving — reflecting how effective AI systems are built in practice: cross-functional, iterative, and execution-focused. Recognition at the IBM TechXchange Hackathon reinforces our commitment to building responsible, enterprise-grade AI solutions that can scale across industries and use cases. While this initiative focused on disaster response, the same agentic patterns are highly relevant to operations, supply chain, manufacturing, and customer support environments.

We look forward to continuing our exploration of orchestrated AI systems that help organizations — and individuals — respond faster, make better decisions, and operate with greater resilience.