Massimiliano Pianges, Founder and COOThe company combines digital product innovation and industry application expertise to develop value-added, customized solutions that facilitate total business transformation. The innovative solution suite enables numerous telco, utilities, banking, insurance, and manufacturing companies to systematize a hybrid cloud infrastructure to attain effective digital transformation— all with increased scalability and rapid cost optimisation.
A Boutique Approach to Cloud Migration
Nexum assists companies in realizing their digital transformation objectives and propels business growth through a wide-ranging service portfolio that includes software engineering, server-less cloud migrations, and applied AI/ML solution development. “By leveragingGoogle Cloud’s data analytics, ML tools, containers and other cloud-native applications, we have developed some leading-edge solutions that are streamlining and optimizing our client’s technology infrastructure while effortlessly facilitating company-wide digital transformation,” says Massimiliano Pianges, Founder and COO, Nexum. Capitalizing on its Google Cloud expertise, the company has developed prototypes of cognitive assistants meant to assist users in creating full cloud infrastructures with reduced development cycles and faster time to market.
One of Nexum’s key proprietary solutions for building cloud infrastructure is Nexcode, an autoregressive language model that uses deep learning algorithms to produce human-like text. Built with scripts based on the new version of Generative Pre-Trained Transformer 3 (GPT-3) and BERT framework, the model runs on a neural recommendation engine based on NLP that uses concepts from computer science, artificial intelligence, and linguistics to analyze natural language, to derive meaningful and valuable information from text. The highly sophisticated and engaging machine learning application is based on the concepts of information extraction (IE)—analyzing text, decomposing it and identifying semantically defined entities and relationships within it to make the text’s semantic structure explicit. Subsequently, whenever users input some text as a “prompt,” the model will generate a text completion thatattempts to match the user’s context or pattern. This empowers a generic user without substantive technical knowledge to quickly develop a software stack application for the whole cloud infrastructure and effortlessly migrate to the cloud.
Next in line is Nexbot, an advanced cognitive assistant that allows users to design and set up a cloud infrastructure using the natural sketch language. Deployable in various settings according to customers’ requirements, it is built to assist any client in constructing a cloud system architecture based on the Google Cloud Platform. The solution is based on a neural network engine designed with BERT that assists users in a wide range of tasks—from designing particular app architecture to calculating expenses of any goods and services by utilizing NPL. It also helps clients adopt serverless architectures and tools to accelerate their platform development in the cloud transition program efforts and restructure legacy systems without prior knowledge of the Google Cloud ecosystem.
The Automation of Automation
Apart from devising solutions empowering successful cloud migration strategies, Nexum is also relying on the power of ML to take information processing to the next level. The company is dedicatedly endorsing ‘automated machine learning’ or simply ‘The automation of automation’. The goal of AutoML is to allow a machine to mimic the way humans design, tune, and encode ML algorithms to help organizations adopt ML more easily. The company follows a three-pronged approach to AutoML, beginning with an optimized search space that comprises a set of hyperparameters and the ranges of each hyperparameter to be selected from. The search space can be a pool of ML algorithms or be the hyperparameters of a specific ML algorithm, such as the structure of the ML model as shown in the pseudocode. Nexum has also designed the search space to be highly task-dependent and adaptable to different ML algorithms for various tasks. It is personalized and ad hoc, depending on the user’s interests, expertise, and experience level.
Since AutoML is often an iterative trial-and-error process, the search strategy to select the optimal set of hyperparameters from the search space usually sequentially selects the hyperparameters and evaluates their performance. It may loop through all the hyperparameters in the search space (as in the pseudocode), or the strategy may be adapted based on the already evaluated hyperparameters to increase the efficiency of later trials. On that account, Nexum believes in adopting a far more advanced search strategy that can help clients formulate a better ML solution within the same amount of time and allow them to use a larger search space byslackening the search time and computational cost.
The third and most essential component of Nexum’s AutoML strategy is performance evaluation premised on a specialized methodology to evaluate the performance of a specific ML algorithm instantiated by the selected hyperparameters. The evaluation criteria are often the same as those used in manual tuning, for example, the validation performance of the model learned from the selected ML algorithm. “Our unique, methodological approach to machine learning is empowering companies to drastically reduce operational costs, increase work efficiency, ensure round the clock data management to accurately predict errors or fraud and take proactive actions in the nick of time,” adds Pianges.
NexumOps
As a company that acknowledges the truth, ‘collaboration is the key to success in software development’, Nexum puts the much needed emphasis on DevOps as a service. In this regard, the company has implemented NexumOps both internally and externally for customers. Each process is completely automated with all the required tools provisioned and configured on request, without any manual intervention.
Upon kickstarting a new project, the Developers/PMs at Nexum compile only a Google Form using standardized templates and rules as each project is created automatically. Thereon, the DevOps team is immediately informed and provided with the Terraform code used to provision each resource for future maintenance, while project creation requests are withheld for further analysis and monitoring.
In parallel, Nexum operates on a cloud native paradigm for building applications as microservices, asset or any kind of file, and running them on containerized and dynamically orchestrated platforms that fully exploit the advantage of the cloud computing model. These applications are developed using the language and framework best suited for the functionality. They’re designed as loosely coupled systems, optimized for cloud scale and performance, use managed services, and take advantage of continuous delivery to maximize the speed, scalability, and finally, profit margin.
In this scenario, Nexum exemplifies continuous integration (CI). In a classic CI pipeline, it triggers a build whenever a code commit occurs and runs the unit tests and all pre-integration tests (quality and security tests). Subsequently, the artifact is built (docker image, zip file, machine learning training model) and is run through acceptance tests while the results are pushed to an artifact-management repository such as a Docker Registry, Cloud Storage, Sonatype’s Nexus, or JFrog Artifactory.
Apart from devising solutions empowering successful cloud migration strategies, Nexum is also relying on the power of ML to take information processing to the next level. The company is dedicatedly endorsing ‘automated machine learning’ or simply ‘The automation of automation’. The goal of AutoML is to allow a machine to mimic the way humans design, tune, and encode ML algorithms to help organizations adopt ML more easily. The company follows a three-pronged approach to AutoML, beginning with an optimized search space that comprises a set of hyperparameters and the ranges of each hyperparameter to be selected from. The search space can be a pool of ML algorithms or be the hyperparameters of a specific ML algorithm, such as the structure of the ML model as shown in the pseudocode. Nexum has also designed the search space to be highly task-dependent and adaptable to different ML algorithms for various tasks. It is personalized and ad hoc, depending on the user’s interests, expertise, and experience level.
Since AutoML is often an iterative trial-and-error process, the search strategy to select the optimal set of hyperparameters from the search space usually sequentially selects the hyperparameters and evaluates their performance. It may loop through all the hyperparameters in the search space (as in the pseudocode), or the strategy may be adapted based on the already evaluated hyperparameters to increase the efficiency of later trials. On that account, Nexum believes in adopting a far more advanced search strategy that can help clients formulate a better ML solution within the same amount of time and allow them to use a larger search space byslackening the search time and computational cost.
The third and most essential component of Nexum’s AutoML strategy is performance evaluation premised on a specialized methodology to evaluate the performance of a specific ML algorithm instantiated by the selected hyperparameters. The evaluation criteria are often the same as those used in manual tuning, for example, the validation performance of the model learned from the selected ML algorithm. “Our unique, methodological approach to machine learning is empowering companies to drastically reduce operational costs, increase work efficiency, ensure round the clock data management to accurately predict errors or fraud and take proactive actions in the nick of time,” adds Pianges.
NexumOps
As a company that acknowledges the truth, ‘collaboration is the key to success in software development’, Nexum puts the much needed emphasis on DevOps as a service. In this regard, the company has implemented NexumOps both internally and externally for customers. Each process is completely automated with all the required tools provisioned and configured on request, without any manual intervention.
Upon kickstarting a new project, the Developers/PMs at Nexum compile only a Google Form using standardized templates and rules as each project is created automatically. Thereon, the DevOps team is immediately informed and provided with the Terraform code used to provision each resource for future maintenance, while project creation requests are withheld for further analysis and monitoring.
In parallel, Nexum operates on a cloud native paradigm for building applications as microservices, asset or any kind of file, and running them on containerized and dynamically orchestrated platforms that fully exploit the advantage of the cloud computing model. These applications are developed using the language and framework best suited for the functionality. They’re designed as loosely coupled systems, optimized for cloud scale and performance, use managed services, and take advantage of continuous delivery to maximize the speed, scalability, and finally, profit margin.
In this scenario, Nexum exemplifies continuous integration (CI). In a classic CI pipeline, it triggers a build whenever a code commit occurs and runs the unit tests and all pre-integration tests (quality and security tests). Subsequently, the artifact is built (docker image, zip file, machine learning training model) and is run through acceptance tests while the results are pushed to an artifact-management repository such as a Docker Registry, Cloud Storage, Sonatype’s Nexus, or JFrog Artifactory.
This practice of having a shared and centralizedcode repository helps developers in directing all changes and features through a complex pipeline before integrating them into the central repository (such as GitHub, Bitbucket, or GitLab).
Walking the Extra Mile
Going beyond the realm of software implementation, configuration, and testing that substantiates the smooth completion of digital transformation projects; the company is also adding quantum computing to its portfolio. They have already started using TensorFlow Quantum (TFQ), an open-source framework that supports high-performance quantum circuit simulators and provides high-level abstractions for developing discriminative and generative quantum models in TensorFlow. At the same time, the company is also experimenting with different theories related to quantum-classical hybrid neural networks through illustrations of TFQ functionalities in several basic applications, such as quantum control, simulation of noisy quantum circuits, and approximate quantum optimisation. Together with quantum neural networks that are helping create novel information systems, Nexum is also exploring quantum systems based on programmable photonic circuits as they also are helpful in creating efficient quantum deep learning networks.
“Our unique, methodological approach to machine learning is empowering companies to drastically reduce operational costs, increase work efficiency, ensure round the clock data management to accurately predict errors or fraud and take proactive actions in the nick of time”
What’s more, Nexum has been using a cross-platform Python library for differentiable programming of quantum computers called PennyLane. It is an open source software framework built around quantum differentiable programming to achieve machine learning tasks with quantum computers. It seamlessly integrates classical machine learning libraries with quantum simulators and hardware, giving users the power to train quantum circuits. PennyLane, while being QMLcapable, is also equipped with several versatile features—the reason why it can support hybrid quantum and classical models allowing users to connect quantum hardware with PyTorch, TensorFlow, and NumPy, thereby boosting quantum machine learning capabilities. In light of these developments, an R&D department on Quantum Programming has been active in Nexum since 2020 that constantly shares breakthroughs in quantum programming and the latest industry insights with customers.
Last but not least, as a company known for having a passion for continuous innovation, Nexum believes in encouraging a dynamic workforce that is always looking to make the best use of the latest technological innovations and strengthen their capabilities. Accordingly, the company regularly conducts a special program named Continuous Improvement. It aims to enhance an employee’s technical and soft skill abilities through various tools and authorize them to attain Google professional certifications to widen their knowledge base and technical know-how of Google Cloud solutions.
In tandem, the company has been developing a serverless Nexum AI-based Cloud Adoption Platform to assist businesses in negating complexities associated with managing an IT infrastructure. Since serverless platforms also remove the need for complex planning or auto-scaling setups, Nexum is planning to build them so that non-AI experts can use them, thereby democratizing access to AI technology and expediting consumption of AI-as-a-Service. “We believe that technologies like Robotic Software Process Automation and quantum programming have been game-changers across the global IT landscape, and we want to be one of the key players in that game!” states Pianges.
By leveraging Google Cloud’s data analytics, ML tools, containers and other cloud-native applications, we have developed some leading-edge solutions that are streamlining and optimizing their technology infrastructure while effortlessly facilitating company-wide digital transformation
Going beyond the realm of software implementation, configuration, and testing that substantiates the smooth completion of digital transformation projects; the company is also adding quantum computing to its portfolio. They have already started using TensorFlow Quantum (TFQ), an open-source framework that supports high-performance quantum circuit simulators and provides high-level abstractions for developing discriminative and generative quantum models in TensorFlow. At the same time, the company is also experimenting with different theories related to quantum-classical hybrid neural networks through illustrations of TFQ functionalities in several basic applications, such as quantum control, simulation of noisy quantum circuits, and approximate quantum optimisation. Together with quantum neural networks that are helping create novel information systems, Nexum is also exploring quantum systems based on programmable photonic circuits as they also are helpful in creating efficient quantum deep learning networks.
“Our unique, methodological approach to machine learning is empowering companies to drastically reduce operational costs, increase work efficiency, ensure round the clock data management to accurately predict errors or fraud and take proactive actions in the nick of time”
What’s more, Nexum has been using a cross-platform Python library for differentiable programming of quantum computers called PennyLane. It is an open source software framework built around quantum differentiable programming to achieve machine learning tasks with quantum computers. It seamlessly integrates classical machine learning libraries with quantum simulators and hardware, giving users the power to train quantum circuits. PennyLane, while being QMLcapable, is also equipped with several versatile features—the reason why it can support hybrid quantum and classical models allowing users to connect quantum hardware with PyTorch, TensorFlow, and NumPy, thereby boosting quantum machine learning capabilities. In light of these developments, an R&D department on Quantum Programming has been active in Nexum since 2020 that constantly shares breakthroughs in quantum programming and the latest industry insights with customers.
Last but not least, as a company known for having a passion for continuous innovation, Nexum believes in encouraging a dynamic workforce that is always looking to make the best use of the latest technological innovations and strengthen their capabilities. Accordingly, the company regularly conducts a special program named Continuous Improvement. It aims to enhance an employee’s technical and soft skill abilities through various tools and authorize them to attain Google professional certifications to widen their knowledge base and technical know-how of Google Cloud solutions.
In tandem, the company has been developing a serverless Nexum AI-based Cloud Adoption Platform to assist businesses in negating complexities associated with managing an IT infrastructure. Since serverless platforms also remove the need for complex planning or auto-scaling setups, Nexum is planning to build them so that non-AI experts can use them, thereby democratizing access to AI technology and expediting consumption of AI-as-a-Service. “We believe that technologies like Robotic Software Process Automation and quantum programming have been game-changers across the global IT landscape, and we want to be one of the key players in that game!” states Pianges.
× 
CIO Applications Europe Weekly Brief
Be first to read the latest tech news, Industry Leader's Insights, and CIO interviews of medium and large enterprises exclusively from CIO Applications Europe
