Projects

DISCO is a dependently typed string and runtime LLVM compiler pass that detects string formatting issues, such as typos. Whenever a string deviates from the expected pattern, the runtime will identify this as a mismatch. This helps reduce usability issues arising from cultural variations in input and output validation.

Here is the paper with more details.

In this project, we developed a parallel implementation of the find command using OpenMP. We also implemented a sequential version of the find command for comparison purposes, benchmarked our parallel implementation against the sequential one, and evaluated the performance difference.

Here is the paper and presentation slides with more details.

In this project, we developed a tool for static reachability analysis using the SVF framework and LLVM IR. We analyze LLVM IR files to statically determine if a function call src() can reach another function call sink(). We generate an interprocedural control-flow graph (ICFG) of a program, which maps control instructions from program entry to exit, and use it for eliminating dead code.

In my undergraduate thesis, I investigated various methods for effectively addressing the Twin-Width problem. I created several heuristics along with a dedicated solver, conducted performance benchmarks on them, and assessed the outcomes by examining how these heuristics performed across diverse graph types.

Here are the slides from the defense presentation.

As part of the Distributed System Prototyping class at TU Berlin, we developed a prototype for a federated marketplace. This is akin to Mastodon but in the Amazon world. I developed multiple components allowing shops based on microservice architecture to easily integrate our solution and collaborate with other shops.

Here are the slides from the final presentation.

This project is a collection of algorithms and heuristics for solving the Vertex Cover problem. We explored and implemented state-of-the-art approaches of solving vertex cover, including reduction rules, branch-and-reduce algorithm with packing constraints and automated parameter optimization techniques with parallel algorithm configuration.

This project was implemented within two weeks as part of the application for the GenDev scholarship for talented computer science students. It is a Telegram bot that enables users to find and directly book package tours within the messenger. The project includes optimized search over a large dataset (15GB+, 100+ million entries), payment integration, and pagination development.

This Telegram bot represents the module catalogue at the Technical University of Berlin. Since finding any module required more than 5 clicks, I decided to port it to the messenger to make life easier. This project includes the implementation of a parser, pagination, and a custom review system for the classes.