Mert Yalhi

An IoT platform that serves multiple IoT products owned by different manufacturers. I designed and developed the platform in only three months. It replaced the previous IoT back end of Nuvia, a Turkish IoT company that provides smart energy solutions.

The previous IoT back end had bottlenecks, which I solved. Still, it did not allow us to integrate new IoT hardware with different communication protocols and couldn't scale horizontally. We decided to build a new platform that replaced the previous IoT back end and migrated all devices to the new platform without any downtime, a seamless transition.

The new IoT platform can scale to support an infinite number of devices and process an unlimited number of telemetries. With a completely similar infrastructure to the previous back end, it can process five times more telemetries. By building this new platform, we've allowed the onboarding of four new companies and increased our device count to more than 100,000 by hosting their devices in a couple of months. It currently serves over 100,000 IoT devices and processes over 10 million telemetries per hour. It also provides real-time data to tenants and useful insights to end users based on the underlying analytics services.

Designed and built an ML-backed decision support system for the customer support department that served over 50 brands and 500,000 customers. The system supported around 200 support agents by providing preliminary analysis on the customer issue and suggested possible solutions and linked issues.

System features:
• NLP-based input analysis that analyzes customer requests.
• Classification of the customer support request based on the predefined workflows and KB articles.
• Clustering of the support cases for a product, which provides insights to the product owners for improving the product and service.
• Automated SEV 1 issue discovery, which detected SEV 1 issues based on customer requests and automatically escalated them to the L2 and L3 agents and product team.

Designed and built an ML-backed intelligent ticket routing system that routed incoming support tickets to the support agents based on the ticket content, previous customer interactions, and support agent workload.

Features and results:
• Routed tickets to a preferable support agent by analyzing the requested content with NLP and matching agent skills.
• Reduced support agent workload by routing the tickets based on agent workload.
• Reduced initial response time to less than minutes.
• Reduced average resolution time to less than 45 minutes.
• Increased one-touch resolution by 60%.

I built an analytics platform that measured and reported customer support and support agent metrics. The system integrated the support system (Zendesk), employee monitoring system, and several other products.

Features:
• Real-time integration to Zendesk.
• Integration to the employee monitoring system for reporting support agent performance metrics.
• Integration to the supported products for providing support-related metrics of the products.
• Provided personalized support metrics to the support agents.
• Provided product-based support metrics to the product owners.
• Provided support department health metrics to the management.