John Wesey Subas

• Optimized slow Apache Spark-based data processing jobs for multiple teams, achieving a 50% reduction in resource utilization.
• Designed and deployed an end-to-end MLOps framework, improving the Mean Time to Reliable Insights (MTRI) and enabling seamless troubleshooting in production.
• Established OCI Data Science as the standard for ML workflows across GBUs, enhancing accessibility and standardization.
• Debugged and resolved performance bottlenecks, such as incompatibilities in CPU instruction sets affecting analytics pipelines.
• Led a technical excellence forum to set direction for evolving data and ML processing pipelines.

Optimized analytics jobs using vectorization in Pandas and NumPy, reducing query processing time by 50% and addressing memory usage and performance issues.

• Improved ML model performance by addressing ensemble model weaknesses and applying hyperparameter tuning, K-fold cross-validation, and other optimization techniques.

• Developed a Python library to auto-generate HTML model cards for various use cases.

• Backend/ML developer and architect at Guavus with expertise in Kafka, ElasticSearch, Postgres, HBase, Apache Spark, Kubernetes, and Spring Boot.
• Led the EDA of NTT Alarms data, driving the analytics team to create models with 90% accuracy and 80% precision/recall for alarm prediction.
• Spearheaded Scala-based ML data pipelines for AlarmIQ and OPSIQ, enabling network service providers to monitor and predict alarms and user experience degradation, handling 5K+ records per second.
• Developed a Scala-based load testing application using Gatling, integrated with Jenkins, supporting KeyCloak authentication and custom field generation, to be used across teams for testing.

• Developed a management VM on Citrix Xen hypervisor for centralized control of all VMs in a Netscaler/CloudBridge instance.

• As a DOCSIS 3.0 expert and C++ developer, I designed the multicast control plane and led the implementation of the cable modem registration FSM for Motorola CCAPs.

• Diagnosed and resolved complex memory-related issues, including memory corruption, segmentation faults, and memory leaks, in Cisco CMTS software.
• Contributed to the stability and performance of critical networking features by optimizing memory usage and ensuring robust system behavior.
• Developed and implemented the DOCSIS Multicast QoS feature for Cisco’s MC2020 linecards with efficient control-plane and data-plane implementations.
• Fixed numerous complex bugs in Cisco CMTS, including customer-reported issues, earning multiple awards for contributions.