Mumbai Flood

1   Background and motivation

Mumbai Flood was developed through a CSR collaboration between HDFC ERGO and the IIT Bombay Climate Center. The objective was to make flood-related information available early enough to be useful, bringing forecasts and location-based observations into a public interface. I led technical development within a team of more than 20 people, coordinating the application components needed to turn incoming data into a usable monitoring platform.

2   Application and data pipeline

The platform used a React frontend, a Django backend, and PostgreSQL for forecasts, location information, and alerts. The frontend presented the changing information, while the backend handled the data and operations needed to support it. Keeping location records alongside forecasts allowed the interface to connect a prediction with a place rather than show it as an isolated model output.

Weather forecast data from GFS fed the rainfall and flood-risk modelling work. The system also incorporated signals from Twitter as a way to surface reports from the ground. These inputs have different roles: a forecast estimates future conditions, whereas a social report may describe something already occurring. Combining their presentation requires keeping that distinction understandable.

3   Development responsibilities

Leading the technical work meant connecting the model and data pipeline to the full-stack application rather than delivering only a prediction script. Forecast storage, alert records, and the public interface needed compatible representations of time and location. The application also had to support updates as information changed, since a static display would quickly become less useful for monitoring.

The project brought together engineering and domain work across a large team. Its value depended on the complete route from source data to an interpretable view, including the availability of the service and the freshness of its inputs. Those operational concerns were part of the platform’s usefulness alongside any individual modelling component.

4   Outcome and maintenance

The project record reports more than 10,000 weekly users at peak usage. The platform is now maintained by another team, so the report describes my contribution during its development rather than asserting ownership of its current implementation. Forecast uncertainty and incomplete ground reports remain important when interpreting this kind of system; a monitoring interface helps communicate available evidence but cannot make every local condition observable.