Engineer at the Memory Solutions Lab working on value proposition of storage solutions for SOTA Deep Learning Algorithms:

• Developed a Verilog code generation assistant with a chat-based UI using various code-gen LLMs, with the best performing model achieving 80% pass@5 success rate in Verilog code generation, benchmarked with VerilogEval, enhancing user interaction and development efficiency
• Designed a movie recommendation demo for DLRM inference utilizing Memory Semantic SSD and showcased the same in the Open Compute Project conference.
• Established a Jenkins pipeline to automate performance tests and also created a visualization dashboard and streamlined performance testing. Accelerated performance testing by 10% through the automation.

Fulfilled the role of a TA for a reputed Deep Learning course (11-785) at CMU, taught by Prof. Bhiksha Raj. Responsibilities include:

• Creating Assignments and quizzes.
• Holding Office Hours.
• Project group mentoring

Worked on ensuring safety of shared control in autonomous driving under the guidance of Prof. Corina Pasareanu.

• Performed K-means clustering on driver reaction time classification output of a neural network to verify robustness of classification and ensure safety of shared
  control in autonomous driving.
• Improved the clustering methodology through mean centroid initialization and elbow method to observe a maximum of 20% increase in cluster radii across 5 clustered regions.

Fulfilled the following responsibilities:

• Trained Feature Pyramid Networks for text localization in documents.
• Enhanced text localization accuracy by 2% after adopting Progressive Scale Expansion Network architecture.
• Resolved issue of omission of isolated characters by tuning hyperparameters for a ResNet-50 backbone.