Epidemic Sentinel Agent

Inspiration

COVID-19 revealed major gaps in India’s public health response due to fragmented data and delayed reporting. A proactive AI sentinel can monitor urban and rural regions in real time, detect early outbreak signals, optimize scarce medical resources, analyze multilingual data, and deliver targeted public health messages. This enables faster, data-driven interventions, helping authorities act before outbreaks escalate and improving outcomes across India’s diverse population.

What it does

Real-time Data Surveillance: Scrapes and analyzes social media, news, and medical reports to flag early signs of outbreaks.

Multilingual Analysis: Leverages Generative AI to process unstructured text in India’s diverse languages, making the system inclusive and regionally relevant.

Proactive Threat Prediction: Uses predictive models to map potential cases on a geospatial grid and forecast disease spread, identifying high-risk “hot zones.”

Automated Resource Allocation: Translates predictive insights into logistics—suggesting optimized distribution of medical supplies and mobile health teams across regions.

Centralized Workflow Orchestration: Powered by IBM’s ADK, ensuring smooth handoffs between specialized agents.

Automated Public Communication: Generates culturally tailored, multilingual health advisories to combat misinformation and build trust.

Paradigm Shift: Moves public health management from reactive to proactive with real-time intelligence and autonomous response capabilities.

How we built it

Surveillance Agent: Python scripts to collect and preprocess data.

Prediction Agent: IBM Granite model + predictive models for outbreak forecasting.

Resource Allocation Agent: Optimization algorithms for logistics.

Community Engagement Agent: Multilingual AI for public messaging.

ADK Orchestration: Connects and manages all agents seamlessly.

Challenges we ran into

A major challenge will be handling the sheer volume and diversity of real-world data. We expect hurdles in sourcing reliable, real-time data from social media due to API rate limits and the highly unstructured, ambiguous nature of such information. To address this, we may build a mock data generation system to simulate realistic conditions. Another anticipated challenge is achieving high accuracy in multilingual NLP, especially when dealing with informal language and regional dialects. Demonstrating a full, end-to-end autonomous workflow within a hackathon’s time constraints will also be difficult, so we will aim to carefully plan a streamlined demonstration of each agent’s functionality and their interactions.

Accomplishments that we're proud of

We aim to create a cohesive, end-to-end agentic workflow that turns pandemic response from reactive to proactive. Our goal is to design seamless handoffs between specialized agents—from data ingestion to predictive analysis and automated communication. In particular, we hope to highlight the Surveillance Agent’s ability to process unstructured, multilingual data, which is crucial for diverse regions like India. Another milestone will be showcasing how the Resource Allocation Agent can autonomously suggest logistical solutions, demonstrating the decision-making potential of an AI-powered system.

What we learned

We learned that the true power of Generative AI in real-world applications lies in its integration within an agentic framework. A single powerful model is not enough; it needs to be part of a well-orchestrated system. The ADK taught us the principles of agentic design, proving that breaking down a complex problem into smaller, specialized agents leads to a more robust and scalable solution. We also learned that effective problem-solving in this domain requires a deep understanding of not just the technology, but also the real-world operational challenges, such as data quality, latency, and the ethical considerations of autonomous decision-making.

What's next for Epidemic Sentinel Agent

After the initial prototype, we plan to integrate the system with real-world public health dashboards and hospital APIs, moving beyond mock data streams. We will explore more advanced predictive models that include environmental and climate variables to improve forecasting accuracy. Another priority will be building an intuitive dashboard for public health officials, enabling visualization of agent insights and manual overrides when needed. Finally, we envision a more robust, conversational Community Engagement Agent capable of answering public queries in real time, becoming a trusted source of information and an active tool to combat misinformation at scale.

Built With

• custom-predictive-models
• ibm-data-preprocessing-toolkits-ai-models-&-apis:-ibm-granite-models
• languages:-python-frameworks-&-platforms:-ibm-agent-development-kit-(adk)
• mock-apis-databases:-postgresql-(structured-data)
• mongodb-(unstructured-data)-cloud:-ibm-cloud-for-deployment
• news-&-social-media-apis
• scaling