Inspiration

In north-eastern Bangladesh, particularly the Sylhet–Sunamganj haor region, flooding is more than an environmental challenge. It can disrupt transportation, damage homes and crops, contaminate drinking water, interrupt access to healthcare and increase health risks for already vulnerable communities.

The June 2022 floods showed how severe this challenge can become. The India Meteorological Department recorded 972 mm of rainfall at Cherrapunji in 24 hours during June 2022. The International Federation of Red Cross and Red Crescent Societies reported that approximately 94% of Sunamganj town and 84% of Sylhet district were submerged, with an estimated 7.2 million people affected across nine north-eastern districts.

These events inspired a central question:

How can we connect flood information with practical community health-preparedness decisions before conditions become more difficult?

Existing flood information is often presented through rainfall measurements, water-level charts or warning announcements. However, communities also need to think about clean drinking water, essential medical supplies, shelter readiness, healthcare access and family safety.

That is the problem FloodGuard AI was designed to explore.

What it does

FloodGuard AI is a bilingual, full-stack flood-health preparedness and decision-support prototype designed around the challenges faced by flood-prone communities in Bangladesh.

Rather than focusing only on flood measurements, it connects flood context with health preparedness and response-planning workflows.

Its main capabilities include:

  • Flood Situation Dashboard: Presents precipitation context, modelled river-discharge estimates, station reference levels and flood-related indicators in one interface.
  • Flood Health Watch: Demonstrates transparent, assumption-based preparedness indicators for cholera, typhoid, skin infections and snakebite-related risks.
  • Shelter Medical Readiness: Helps explore sample inventories of oral rehydration salts (ORS), clean drinking water and first-aid supplies.
  • AI Flood Simulator: Allows users to explore hypothetical rainfall and upstream-influx scenarios.
  • RouteSafe: Demonstrates how road accessibility, shelter locations and healthcare access could be considered during flood-response planning.
  • Interactive GIS Map: Uses Leaflet to present geographic context, station references, shelter locations and other mapped information.
  • Community Ground-Truth: Allows people to submit flood-depth and road-condition reports with consent. Submissions remain pending until reviewed by an authorized administrator.
  • Operations Center: Includes check-ins, alert rules, event updates, responder workflows, audit records and a situation-report generator.
  • Family Safety Plan: Provides household preparedness checklists, emergency contacts and a shareable safety plan.
  • Bilingual Experience: Supports both English and Bangla to make the interface more accessible to local communities.

The project also distinguishes live API information, modelled estimates, calculated indicators, community-submitted information and simulated examples.

Important limitation: FloodGuard AI is a research and demonstration prototype. It is not an official flood-warning service. Some station levels, shelter inventories, road conditions and scenario records are simulated or illustrative. Its health indicators are not clinically validated, and its route examples are not verified evacuation instructions.

How we built it

I developed FloodGuard AI as a full-stack web application, combining a modern frontend, a backend API, data-processing utilities and selected AI-powered features.

Technology stack

  • Frontend: React 19, TypeScript, Vite and Tailwind CSS.
  • Maps and visualizations: Leaflet, Recharts and Lucide icons.
  • Backend: Node.js and Express.js.
  • Database and persistence: SQLite for operational records, with JSON-based persistence for selected community reports and public alerts.
  • AI integration: Google Gemini API for selected conversational and natural-language explanation features.
  • Testing: Vitest and Node.js native test runner.
  • Resilience: Progressive Web App support and service-worker caching for previously loaded application screens.

How the system works

The frontend presents flood context, maps, charts, preparedness indicators and response-planning tools. The Express backend handles API requests, data validation, selected AI integrations, report moderation and operational workflows.

Open-Meteo provides precipitation information when available, while the Open-Meteo Flood API provides modelled GloFAS discharge estimates. These are different types of data: modelled discharge is not the same as a direct river-gauge measurement.

The application uses deterministic calculations and statistical methods for its numerical indicators. Google Gemini is optional and supports selected natural-language interactions, such as conversational assistance and explanations. Safety-critical health guidance remains predefined bilingual content rather than being generated by an AI model.

I also implemented validation, rate limiting, consent requirements and moderation workflows to make community reporting more responsible.

Challenges we ran into

One of the biggest challenges was working with different levels of data reliability.

Rainfall observations, gridded hydrological estimates, historical references, simulated station values and community reports cannot be treated as equally reliable measurements. Combining them without clear labels could create a misleading picture of flood conditions.

Another challenge was translating flood information into useful health-preparedness indicators. Reliable disease-outbreak prediction requires appropriate surveillance data and scientific validation. Since this prototype does not have such a validated dataset, I used transparent, configurable planning heuristics rather than presenting the results as clinical predictions.

AI safety was another important consideration. A generative model can produce convincing but incorrect information. To reduce that risk, numerical calculations remain separate from generated text, and critical health-preparedness instructions use fixed guidance.

I also considered the practical limitations of connectivity during emergencies. This led to a bilingual interface, fallback behaviour and service-worker caching for previously loaded screens.

Finally, integrating maps, scenario simulations, community reporting, operational workflows and database-backed features into one application required careful organization and testing.

Accomplishments that we're proud of

I am particularly proud that FloodGuard AI goes beyond a conventional flood dashboard and explores the relationship between environmental conditions and community health preparedness.

Some of the most meaningful accomplishments include:

  • Building a consolidated full-stack application with 20 feature views.
  • Connecting flood context with shelter readiness, health-preparedness indicators and clinic-access scenarios.
  • Implementing a consent-based community reporting workflow with server-side validation and moderation.
  • Developing a SQLite-backed Operations Center with check-ins, event updates, alert rules and audit records.
  • Supporting English and Bangla throughout important user-facing workflows.
  • Adding a Family Safety Plan and accessibility options.
  • Making data provenance visible so that users can distinguish live information from modelled, derived and simulated information.

The consolidated source code also passed automated CI checks, including TypeScript validation, 110 frontend/unit tests, 16 server tests and a production build.

Most importantly, I am proud of the project's focus on transparency. A demonstration value should never be mistaken for an official measurement, and a prototype should never be presented as a proven emergency-warning system.

What we learned

Building FloodGuard AI taught me that developing a meaningful disaster-response application requires much more than writing code or integrating an AI model.

I learned how important data provenance, uncertainty and responsible communication are when working on problems that can affect people's safety.

I also learned that AI is most useful when it supports a well-designed system rather than replacing reliable data, transparent calculations or professional judgement.

Working across React, TypeScript, Express, SQLite, mapping libraries, external APIs and automated testing strengthened my full-stack development and system-design skills.

Most importantly, I learned to think about the people who might eventually use a tool like this: a family preparing for flooding, a volunteer checking local conditions or a responder organizing essential supplies.

What's next for FloodGuard AI

FloodGuard AI is an ongoing prototype, and several improvements are planned:

  1. Verified local data integration: Explore reliable integration with Bangladesh Water Development Board (BWDB) gauge data and official Flood Forecasting and Warning Centre (FFWC) bulletins, subject to availability and access requirements.
  2. Scientific validation: Compare statistical estimates and planning indicators with appropriate historical observations and publish evaluation results, uncertainty and limitations.
  3. Community testing: Conduct usability research with flood-affected residents, volunteers and local organizations before considering real-world deployment.
  4. Improved health preparedness: Work with qualified health professionals to review preparedness content and evaluate whether reliable local health data can support more meaningful indicators.
  5. Low-connectivity improvements: Improve offline usability, data synchronization and access to essential preparedness information.
  6. Responsible notifications: Explore integration with a compatible SMS gateway and verified alert sources, with appropriate consent and delivery safeguards.

The long-term vision is to develop FloodGuard AI into a more reliable, locally informed flood-health preparedness platform that helps communities organize information and prepare more effectively.

FloodGuard AI — Predict Early. Alert Faster. Protect Lives.

Built with a focus on community preparedness, responsible AI and transparent data.

Built With

Share this project:

Updates

Submission history