Back to page

Machine Learning

With advancements in data science it is possible today to quickly and automatically produce machine learning (ML) models that can analyze higher volumes of more complex data and deliver faster, more accurate results – even on a large scale. By building precise models, an organization has a better chance of identifying profitable opportunities – or avoiding previously unknown risks.

These technological advancements have led to the widespread interest and adoption of machine learning, which finds its application everywhere nowadays – in Natural Language Processing, Cybersecurity, Infrastructure Management, Healthcare Decision Support, Training and Education – the list goes on.

Yet, in spite of the technologies that already exist, many companies are faced with the need for specific ML tools that may not yet exist.  We work alongside our customers to design and build those ML tools. In fact, First Line has formed an organizational structure, business processes and methodologies specifically focused on artificial intelligence and machine learning. The structure includes client experts who will have the opportunity to gain valuable knowledge by participating in agile team activities.

Organizing Engagements with Machine Learning

While machine learning may be a critical element of a system under development and require special skills in the engineering team, in most cases it is only one of many elements of the system overall.

Given that machine learning typically deals with a large amount of data, a typical project includes the following steps:

With the supporting data infrastructure in place, the ML-specific tasks may include:

It is highly likely that the system under development will include other, more traditional elements and components, such as user interface, business logic and persistence layer.

Following the above implementation patterns, we structure the engagement and staff the teams accordingly: a balanced mix of data engineers, ML experts, and other developers working side by side with each other and with the client’s engineering organization.

A few of our representative engagements involving machine learning are described below.

Using Machine Learning for Data Governance of Analytical Assets

Health organizations routinely generate thousands of such analytical artifacts on a daily or weekly basis. With the wide-spread adoption of EHR systems by healthcare institutions, large amounts of data become available for reporting and other analytical insights.

Reliable patient matching is one of the most complex and most important elements in communications between health systems. Matching records to the correct individual is more complicated when patients receive care in multiple settings and when organizations and providers use different systems to share records electronically.

A leading healthcare system in the United States approached First Line Software for a solution. Our team created a searchable catalog containing over 100,000 analytical insights and automated the process to organize and classify them. Machine learning algorithms were built to detect duplicates and similarities in the reports and accurately assign the right metrics and portfolio membership.

A significant amount of already-available reporting data provided the foundation and opportunity to train the algorithms. These processes continue to evolve and improve ‒ with minimal human involvement ‒ as new analytical insights are added.

Upon deployment into production this system has been instrumental in reducing the analytical inventory by nearly 20%, lowering the turnaround time for producing new reports nearly in half, and significantly improving the quality of analytics for the organization.

Computer Vision Solution for a Large Industrial Client

Our client’s live streaming video from CCTV cameras had to be continuously monitored and analyzed to determine whether the staff was wearing mandatory safety devices - such as protective helmets or gas masks - in designated areas. An existing solution was dependent on signals from sensors embedded in the site equipment or movements identified by computer vision algorithms.

There were several specific challenges with this project:

  1. Visual recognition systems are typically trained by observing sets of available videos that replicate live scenarios. No such videos demonstrating staff compliance with mandatory safety equipment guidelines were available.
  2. Real-time video analysis was required.
  3. The system had to analyze and recognize specific movement sequences. For example, the following sequence is considered compliant:
    • The employee places a hard-hat on their head
    • The employee places a hood over the top of the hardhat.

Without analysis of existing videos, such sequence could be interpreted as a violation.

First Line addressed these and other challenges to deliver a solution that fully met the client’s technology requirements and reduced costs associated with safety-related injuries of the staff.

Patient Identity Resolution Using Machine Learning Techniques

Our client which develops registries of rare diseases has a need to receive and process patient data coming from external systems. Only patients that match target demographics and disease criteria qualify to be captured in the registry. Additionally, records coming from external systems have to be correctly matched to patients which are already in the registry.

We have utilized and integrated existing open-source machine learning modules with proprietary algorithms developed and optimized in-house. We trained these components on large sets of synthetic and real-life patient records to achieve desired matching accuracy and included patient identity resolution processes in the existing patient data ingestion pipeline.   

Image Recognition Solution for a Media Company

Our customer asked our engineers to develop an online system that could identify and recognize military equipment in a photograph. In the target implementation, the user of the system uploads a photo, asks the system to identify the equipment and provide a detailed description, including the model and its characteristics.

Algorithms used in the implementation of the system were trained using a collection of more than 300 units of combat vehicles from a museum in Europe - one of the most significant collections of military equipment.

We crowed-sourced an additional learning set by asking community experts to classify suggested military equipment online.

In this project we used transfer learning for convolutional networks to significantly reduce the size of each data set.

The First Line team looks forward to learning about your requirements for machine learning capabilities and the opportunity to deliver a high-performing solution that will help your company achieve its business objectives.

About Us

First Line Software is a premier provider of custom software development, technology enablement services and value-add consulting in big data engineering, digitalization, intellectual integration, industrial Internet, and IoT, digital media and marketing, and enterprise content management as well as healthcare IT.

Headquartered in the US, First Line employs 500+ staff globally. First Line team and company culture is centered around subject matter expertise, technical excellence, consulting capabilities and proven methodologies, with a strong focus on Agile and Intellectual Integration.

The company has been recognized with multiple annual rankings and awards by the International Association of Outsourcing Professionals (IAOP), Global Services, CorporateLiveWire, Insights Success and CNews. We were the first to be awarded the Scrum Capability Medallion by Scrum, Inc. Most recently, research firm Gartner included FirstLine in their first ever Market Guide for Technology Integrators (2014) and the Cool Vendor in Applications Services 2015 Report. We are active members in Object Management Group and Industrial Internet Consortium. FLS is also an EPiServer Premium Solutions Partner.