Architecting a Scalable Cloud Infrastructure for Machine Learning Services
Modernizing Poultry Farms with Internet Connected Feed Level Monitoring
Little Bird Systems approached Lofty Labs to create a customer facing application to allow users to monitor data generated by their innovative audio-based sensors for feed bins in the poultry industry. The team had deployed sensors on multiple farms and were storing data in a database, but had no user interface for analytics, reporting, or even internal oversight. The team also anticipated large scaling challenges with their architecture in the short term future as they prepared a large go-to-market strategy for their IoT (Internet of Things) enabled sensors.
LOFTY'S APPROACH
Little Bird Systems team during a discovery workshop at Lofty Labs Lofty Labs worked with the team at Little Bird Systems to re-architect the pipeline of their sensor data using a standardized API (Application Programming Interface) over the internet, allowing their devices to passively push data into the cloud. The API became the standard interface for retrieving data as well, and Lofty developed an interactive reporting interface to surface insights to Little Bird System’s customers. The API and analytics application support an ACL model that allows for flexible governance of data access where Little Bird's employees and customers can access only the data they are granted, without compromising the confidential data of other customers. LBS On ScreenLittle Bird's IoT infrastructure is now poised for massive scaling and sensors are being rolled out to multiple commercial farms. The API powered dashboard is in use by LBS employees, farm employees, and integration partners to efficiently monitor and replenish feed levels. Native mobile applications are now in development, taking advantage of the API at the architecture's core to provide consistent experience across devices. These applications can be developed at a substantially reduced cost as they leverage a pre-existing cloud architecture and data access. Little Bird Systems is now confidently approaching some of the largest poultry integrators in the world with the knowledge that their automated systems can scale to meet demands of any size
When Mondobrain engaged Lofty they had a powerful predictive analytics algorithm and an opportunity to win France’s largest vaccine producer. Mondobrain needed to wrap their algorithm in a platform that worked around the globe and passed the most stringent of web application security audits. Further, they needed software that could stand up to the use cases of big-pharma laboratories with hundreds of users: their prototype could only predict on one dataset at a time, often taking 5 to 10 minutes to complete a single job.
Lofty Labs consultants worked directly with Mondobrain’s staff an as extension of their team to build world-class software around their algorithm. The algorithm was moved from a single machine to an Amazon Web Services cloud environment wrapped in a web application layer built with the Django framework. The application layer handles the customer facing application, and the delegation of algorithm solve jobs to a scalable pool of worker servers. In addition, Lofty Labs worked with Mondobrain to introduce a cutting edge technology which compiled Mondobrain’s algorithm to extremely fast machine code without changing how the algorithm was written. Leveraging these increased speeds and other open source cloud computing tools, Lofty Labs engineered a scalable application that could handle high volumes of simultaneous predictions.
Mondobrain began a pilot program with their flagship client within two months of engaging Lofty Labs, where the software passed audits from multiple Enterprise IT security firms, including IBM. Lofty’s optimizations to the algorithm saw speed improvements over 50x, and with the ability to distribute solves across a cluster of cloud computers, Mondobrain’s prediction throughput increased over 100x on a three server cluster. The software can now be scaled with the click of a button to add additional server resources. Lofty’s deployment and infrastructure automation tools have enabled Mondobrain to uniquely serve each of the customers they acquired over the following years. The same Mondobrain software is now deployed in Amazon Web Services data centers in the United States and Europe as well as a number of private enterprise networks and physical appliance configurations. Lofty Labs has continued to work with Mondobrain since 2015 as they’ve grown to an 8-figure business and our consultants lead architecture and strategy initiatives within Mondobrain’s internal engineering team.
Distributing the outcomes of data science and AI processes within an organization requires a product development approach. Lofty Labs excels in turning data driven processes into scalable, functional software applications. Software scalability is achieved through execution speed and concurrency. Lofty Labs understands how to optimize software through both code execution speed and infrastructure scalability. Cloud native applications can be designed in a way that allow them to run in on-premise and appliance configurations, but that design must be factored into the lowest architecture of the software. Lofty Labs knows how to design software that leverages the cloud, while maintaining its portability to other systems.
