In this paper, we design and implement an Internet-of-Things (IoT) based platform for developing cities using environmental sensing as driving application with a set of air quality sensors that periodically upload sensor data to the cloud. Ubiquitous and free WiFi access is unavailable in most developing cities; IoT deployments must leverage 3G cellular connections that are expensive and metered. In order to best utilize the limited 3G data plan, we envision two adaptation strategies to drive sensing and sensemaking. The first technique is an infrastructure-level adaptation approach where we adjust sensing intervals of periodic sensors so that the data volume remains bounded within the plan. The second approach is at the information-level where application-specific analytics are deployed on board devices (or the edge) through container technologies (Docker and Kubernetes); the use case focuses on multimedia sensors that process captured raw information to lower volume semantic data that is communicated. This approach is implemented through the EnviroSCALE (Environmental Sensing and Community Alert Network) platform, an inexpensive Raspberry Pi based environmental sensing system that periodically publishes sensor data over a 3G connection with a limited data plan. We outline our deployment experience of EnviroSCALE in Dhaka city, the capital of Bangladesh. For information-level adaptation, we enhanced EnviroSCALE with Docker containers with rich media analytics, along Kubernetes for provisioning IoT devices and deploying the Docker images. To limit data communication overhead, the Docker images are preloaded in the board but a small footprint of analytic code is transferred whenever required. Our experiment results demonstrate the practicality of adaptive sensing and triggering rich sensing analytics via user-specified criteria, even over constrained data connections.