events analyzed / second
MB parsed / second
Telecom – processing an impossible data stream.
LakeTide wrote high performance Julia code to parse and analyze MME performance logs in real-time so that valuable insights could be extracted.
Unlocking Performance Data
Telecom operators often have access to performance log-files in proprietary formats for troubleshooting purposes, but are unable to use them directly because of vendor lock-in. LakeTide reverse engineered a number of file formats and unlocked a treasure trove of valuable information on the performance and overall behavior of the network.
Reducing East-West Traffic
Telecom appliances have a limited ability to scale out for the purpose of real-time analytics and monitoring. This places strict requirements on the physical proximity of servers and how much bandwidth they are allowed to consume for the purpose of transmitting performance data. Using the Julia programming language, LakeTide developed code that could processed all the data on a single 8-core server, in real-time, over a single 1Gb/s NIC.
A.I. powered inference
LakeTide trained an A.I. to reproduce performance metrics which under normal circumstances were only available upon purchasing the hardware vendor’s analytics cluster. Moreover, the AI could do inference in real-time and opened up avenues for new business models using the performance data.
milliseconds of latency
Manufacturing – accelerating digitalization, maximizing uptime.
LakeTide used Azure’s stack of big data and machine learning services to capture billions of IoT events and turn them into data driven services.
From Storage to Actionable Insights
By enabling data that was merely stored in the past, it was possible to train deep neural networks that could react to the dynamics of industrial equipment in real-time. Model predictions could then be fed back both to customers and the analytics team for ongoing product improvements.
Geo-distributed IoT Pipeline
LakeTide implemented a geo-distributed multitenant architecture on Azure that allowed smart IoT devices to send and receive data from across the world. The easily managed SaaS applications allowed developers to focus on putting new services into production and deliver new types of value to customers.
Digitalized Customer Experience
Providing customers with an app where they can track the health of their equipment and get notified in advance when to service parts has significantly enhanced the customer experience. By avoiding unscheduled downtime, the new digital experience helps increases customer satisfaction and cuts costs.
million rows of data
Construction – enabling a data driven organization.
LakeTide used Cloudera Enterprise Data Hub to combine numerous data sources in ways that could be operationalized immediately.
Data from multiple sources such as social media, customer support, finance, project management, devices, and more are all brought together from every country into one place – the data lake. With a single source for analytical insight and user-friendly interactive dashboards, data driven decision making was truly democratized.
Kafka and Spark Streaming are used to process data such as temperature, power use, noise, humidity, and more, from IoT devices in the company’s buildings. In combination with existing operational data, it can be used to provide new innovative services to customers and tenants.
Omnichannel 360° CRM
Customer relationship data is imported into the analytics platform and combined with detailed operational data. This allows the company to track the customer journey all the way from a first facebook-session, to a sale, to long-term support – a truly 360° view. By surfacing the data through powerful interactive real-time dashboards, it empowered the organization to make smarter decisions faster.
GBs of data
Weeks to Completion
FinTech – utilizing AI to analyze images and improve customer services.
LakeTide used GPU accelerated workstations to run machine learning algorithms on a world-class dataset of 20 million+ images and corresponding metadata. The resulting models achieved outstanding performance and catalyzed new forms of product development.
A deep neural network was trained to distinguish between 14000+ extremely similar classes of images. The resulting model reached better-than-human levels of performance at over 43% top-5 accuracy.
A highly custom neural network was designed to detect objects within an image. Existing software could make use of this to ensure that images fulfill important criteria before being sent off for review – significantly decreasing handling time.
A set of additional machine learning models were built to find patterns in company metadata. These models could be used to improve the user experience in existing products by providing intelligent suggestions, detecting anomalies, and much more.
Thank you for a really great engagement! We were impressed by the work you did. Data is the core of our business and we will be capitalizing on the new techniques you’ve introduced.Johan