Hemang

Debugging the debugger (LLDB)

LLDB (an open-source project) is the debugger for the LLVM compiler for the C and C++ programming languages. Over the years, it has been a useful tool for debugging any bugs that programmers may have in their programs. However, it still displays several anomalies when one tries to debug a program compiled in higher optimization levels. For instance, if during optimization, the compiler removes a line from the program, what happens when someone breakpoints to that line in LLDB? This is why I engineered and established a robust infrastructure to systematically identify and diagnose bugs in LLDB's behavioral performance across diverse compilation optimization levels.

Edge-FaaS

Rethinking the way we support their applications is necessary to enable next-generation technologies like self-driving cars and smart cities. The proliferation of devices in the Internet of Things has sped up the development of these technologies. These devices have the potential to produce enormous volumes of data, and the applications that enable them frequently need this data to be handled quickly. In order to properly service these applications, we must complement and extend the Cloud computing model.

For the same, we conducted a research project to show how the Edge could be elevated to be a peer of the Cloud for addressing these challenges. I analyzed workloads to test enabling a FaaS platform to operate efficiently over a geo-distributed continuum of edge clusters, to serve requests and avoid any one edge-cluster from violating latency requirements. I also utilized a SUMO script to generate random activity of vehicles in a city over a period of 36 hours and assigned each vehicle to an edge site closest to it using the Euclidean distance heuristic.

Advancing the performance of GQA

In my freshman year, in search of new research opportunities, I came across this project being done to advance the performance of Graphical Question Answering in the field of Computer Vision. In this project, I enhanced and advanced the performance of Graphical Question Answering by combining GQA dataset with the VQA architecture, which contributed to the building of a new state-of-the-art deep learning architecture using transformer modules.