H. Özgür Çağdaş

I joined the project at the stage where it was being brought in-house from outsourced development and worked at different levels, including:
• Setting up a Jenkins CI/CD system.
• Designing a CI workflow and developing automation and wrapper scripts to execute the system with minimal developer input.
• Integrating the Google Test framework into the system and introducing a unit-tested platform-independent component to the project.
• Developing various components in line with new feature requirements.
• Integrating a basic level of Amazon Alexa in a proof of concept for alarm call triggering.

As well as bug fixing and feature development using C++ and Java, I undertook various data science tasks that involved identifying the most crucial system metrics that impacted the KPIs and then developing predictive ML models. These models were used to identify performance degradation points based on specified features before they became noticeable to the end user, allowing us to take timely corrective action.

I also developed various automation tools using Python to interface with Elasticsearch and SQL-based telemetry and log stats servers, identifying KPI outliers, trend changes, and workaround WebUI query-result-size limitations of these systems.

In addition to the academic background and theory building, this study involved building a machine-learning-based buy and sell signal generator for the AUD/USD forex pair, with a set 1.5% profit target for each position. On top of the data collection automation and data cleanup steps, the project also involved developing an automated feature and method selection based on classification scores and total profit generated. In the end, a tree-based classifier and extra trees were identified as the most optimal method with low training time and profit levels close to the DNN benchmark.