Dedicated Software Development for the Advertising Platform

About the Client

Our client provides an AI-based platform that enables high-precision digital advertising campaigns on YouTube, TikTok, Facebook, and mobile gaming apps. The solution is one of the most audited digital advertising platforms worldwide, accredited by prominent organizations such as COPPA KidSafe, ABC, JicWebs Standards, IAB Gold Standard, etc.

The team, based in London, New York, Sydney, and Sarasota, is a powerhouse of 50+ specialists, including engineers, data scientists, and customer relationship managers. Their expertise and dedication ensure the successful execution of our projects, covering the needs of worldwide clients.

Using context-aware intelligence and cutting-edge ML algorithms, the company helps advertisers analyze millions of videos at a granular level. As a result, advertisers can predict the performance of campaigns before their launch and target the right audiences at the right time.

Goal

The client needed engineering help improving their existing PHP-based advertising platform. They envisioned a new version of the software capable of aggregating attributes of video content, integrating proprietary ML algorithms, and introducing innovative approaches to video ad targeting and contextual advertising.

An updated platform version was expected to have a high-load architecture supporting asynchronous processes and Big Data processing. The solution needed to be prepared for seamless connectivity with Facebook, YouTube, and other platforms and for injecting AI functionality for data analysis and reporting.

Work Description

Duration:

2019- ongoing.

Team:

At the project’s onset in 2019, we assigned a dedicated Java developer to create an MVP prototype, enabling the client’s company to get approval to use Facebook API. The collaboration proved so successful that the client extended our engagement, assigning additional tasks to a Java developer. Additionally, Geomotiv provided senior Python developers to handle back-end development and data engineering tasks.

Services provided:

• Platform migration and improvement (duration – 1.5 years)

Geomotiv provided a senior Java developer who participated in the transition of the client’s legacy software to Java. This was intended to accelerate the solution’s operations, considering the system’s envisioned and present functionality.

One of the project’s main objectives was to automate the extraction of video attributes and statistics from YouTube to find relevant videos and place ads. Our developer added necessary Java libraries and deployed microservices with basic parameters for performing data science tasks.

Geomotiv’s specialist developed a new server application and incorporated microservices into the platform’s architecture. In the course of work, the developer optimized the underlying code, tested the system’s performance, and maintained microservices.

Our Java developer focused on monitoring the back-end infrastructure, ensuring robust data handling and seamless integration with new front-end components. Once the back-end was fully prepared, the client’s front-end development team took over and created a user-friendly interface that aligned with design specifications and UX goals.

• Improvement and expansion of microservices for data science tasks (in progress)

At this stage, a team of data scientists joined the client’s project. They used Amazon’s virtual servers and tools to diversify and expand the use of ML algorithms into the platform.

Our developers enhanced the main microservices, such as Amazon SageMaker, comprising three ML models, Amazon Comprehend, and a client’s custom search service. This successful integration of ML algorithms significantly enhanced the platform’s capabilities, allowing for more accurate predictions and optimized campaigns.

After integrating microservices, the platform was capable of producing comprehensive results. The developers used microservices to upload videos and their attributes, which were automatically transferred to the data science unit for further analysis. The system included over 30 parameters for evaluating videos.

Later in the project, Geomotiv’s Python developers joined the data science team. They helped transform unstructured client data from Excel files and integrate it directly into the Redshift database. Data scientists then extracted relevant datasets from Redshift to generate visualizations and train ML models for consecutive targeting.

• Ongoing development and optimization (in progress)

The client’s team, including our developers, has grown to ten members. Our specialists continue addressing ongoing tasks related to adding new features and maintaining system performance.

Adding Google Play and App Store.

Our developers connected the platform to Google Play and App Store via their APIs by first acquiring the necessary API keys and configuring the development environment. Geomotiv’s specialists then built microservices to interact with these APIs to fetch mobile app data and user metrics, which segment users and create targeted ad campaigns.

Aggregating captions for data analysis.

One of the challenges we faced was the limited support of Google’s API for collecting captions, which made it inconvenient for developers to complete this task. As a result, we had to opt for third-party libraries to handle this functionality. These libraries were not Java-based, which required extra effort to integrate them into the existing platform. However, our team’s problem-solving skills and resilience allowed us to overcome these challenges and deliver a successful solution.

Automation of data parsing.

Geomotiv’s Python developers continuously help improve customer data transformation to deliver accurate results to a data science team. Previously, the company relied on custom Python scripts tailored for each customer to parse datasets that arrived in various formats. This process was both time-consuming and resource-intensive.

Our client entrusted the team to develop a new data parsing service to address these challenges. Our Python developers created a solution that allows developers and non-technical users to configure and manage data parsing in a user-friendly interface. This solution eliminates the need to write custom scripts for each client, reducing the workload on developers by up to 80%. Currently, 85% of the planned functionality of the front- and back-end parts is ready.

Optimization of processes.

The platform is an asynchronous system that requires additional configuration to eliminate potential conflicts. Java provides robust solutions for these issues, including asynchronous processing and transaction support. Our developers leveraged these Java features and various libraries designed for such tasks.

The team also resolves ongoing issues with the platform’s performance. For instance, one element of the system contained multiple tabs that faced a 20-second delay in response during query processing. Our developers split large queries into smaller ones and removed redundant data that slowed performance. Additionally, the specialists manually wrote SQL queries to retrieve only relevant data.

Approach:

• Communication:

The team conducts bi-weekly video meetings with the client’s PM via Google Meet to synchronize the team’s efforts. Internal team communication occurs in Slack to ensure quick resolution of emerging challenges.

• Task management:

The PM assigns tasks for each team member during time-boxed video calls. The PM has an extensive technical background and knows how to set effective task distribution and realistic deadlines. Developers estimate the scope of work and time required for each task and get down to its execution.

• Task review:

The PM reviews completed tasks to comply with high-quality standards and adherence to technical requirements. The team maintains a linear workflow to ensure efficient allocation of tasks and avoid overlaps. Asana is used for transparent task monitoring, tracking progress, and setting priorities.

Work Stages

1. Platform migration to Java
2. Back-end development
3. Integration of ML algorithms
4. Ongoing development and optimization

Results

Geomotiv’s resources have become integral to the client’s engineering team. They successfully transitioned to a high-performance system capable of handling large volumes of data. Our developers optimized the platform’s back-end part and streamlined Big Data processing workflows. The system now processes up to 20 videos (along with all their attributes) per second. Additionally, Big Data pipelines operate seamlessly, with databases capable of storing up to 1 billion videos.