We design, implement and deploy a rural telemedicine network, connecting several villages to the Aravind Eye Hospitals in India, enabling over 100,000 (and counting) remote eye examinations, via high-quality video-conferencing between doctors and remote rural patients who otherwise had no access to any health care services. Of these, close to 20,000 patients have received their sight back, only due to early diagnoses enabled by our network. This dissertation describes the technologies we have created across several layers that enable new options for rural connectivity, more viable than the current state of the art. Long-distance point-to-point links, up to hundreds of kilometers long, are best suited to introduce connectivity into rural geographies, typically characterized by sparsely-spread clusters of lower-income users. Compared to fiber, cellular, satellite and WiMAX technologies, WiFi offers more exciting possibilities as it is very inexpensive, offers operational freedom due to its use of unlicensed frequencies, and is less complex to deploy and manage by rural organizations. However, WiFi performs very poorly in long-distance settings due to low channel utilization, intra-link packet collisions, inter-link interference at relay nodes, and assymetrically lossy channels. In addition, managing systems in rural areas is very challenging due to frequent failures resulting from poor-quality power, operator errors by inexperienced staff, and few possibilities for remote management. Taken together, these challenges prevent financial and operational sustainability, a critical goal rural networks need to achieve to scale beyond small pilot deployments, in order to have any real impact at all. We treat sustainability as a critical systems design goal and take an end-to-end systems perspective by asking the question: How can we design and build financially and operationally sustainable WiFi networks that provide high-throughput in long-distance settings, in the face of lossy environments, and in the presence of sytemic link or node failures? At the network layer, we design and implement WiLDNet, a new TDMA-based MAC-layer that increases link utilization, eliminates most packet collisions in single- and multi-hop settings, and combines FEC and ARQ for link-level loss recovery. Compared to the standard WiFi MAC, WiLDMAC enables 2-5 fold improvements in UDP and TCP throughputs. At the management layer, we build a range of tools for system-monitoring over intermittent links, backchannels for fault diagnosis, and mechanisms for hardware and software failure recovery. At the lowest layer, after careful investigation of the effects of poor-quality grid power, we build low-cost solutions that make wireless nodes more resistant to power-related damage. We also build off-grid power solutions that can extend the life of battery backups, further reducing costs and enabling connectivity in remote areas. These solutions, across all layers, have enabled financial and operational sustainability of the Aravind telemedicine network, resulting in both scaling and replication of our work. Aravind is scaling our network to connect to 50 villages, targeting general access to eye care services for 2.5 million people and 500,000 remote eye examinations annually. Other groups such as the Lumbini Eye Institute, the Pakistan Institute for Community Ophthalmology and Inveneo are using our work to replicate similar networks in Nepal, Pakistan and Uganda respectively.