• Specialization: High Performance Computing

• Relevant coursework: Operating Systems; Computer Networks; Artificial Neural Networks and Deep Learning

• Contributing to enhancements in Quantbot’s backtesting and real-time trading engines, with a focus on evaluating and implementing performance optimizations.

• Developed a Python script to parse and transform batches of internal job logs (~2,000+ lines per run), extracting error patterns and filetype data, reducing 40+ hours of manual log review for a 4-person engineering team.
• Created a PowerBI KPI dashboard to visualize trends in system log errors; used weekly by the team to analyze issues and spot recurring failures.
• Researched and evaluated multiple data confidence platforms (Collibra, Snowflake) to handle terabyte-scale EHR and claims data; summarized findings for the team to inform future data pipeline structure.

• Designed and implemented a multithreaded log analysis tool (4 threads, 3.6× speedup) in C for 10K+ distributed tasks, reducing analysis time by over 70% and enabling faster bottleneck detection.
• Extracted per-task CPU/memory utilization, I/O throughput, and runtime; designed an algorithm to capture idle periods between two tasks in IBM’s DP logs, revealing resource underutilization and scheduling inefficiencies.
• Built a dynamic profiling tool using strace, top, and custom logic to trace CPU, memory, and I/O usage of executing code in real time, enabling detailed performance diagnostics.
• Applied the tool to Synopsys Proteus, analyzed multiple test cases, and identified heavy I/O bottlenecks affecting runtime performance.

• Built a C++ option pricer with OpenMP and CUDA, parallelizing Monte Carlo for European and American (Longstaff–Schwartz) options using real-time market data from Tradier and FRED APIs.
• Achieved 5× CPU / 180× GPU acceleration for European options, and 3× CPU / 47× GPU for American options, versus serial baselines.
• Implemented validation against Black-Scholes and Cox-Ross-Rubinstein models to ensure pricing accuracy.

Code: github.com/brianshen03/MonteCarloOptionPricer

• Built a TensorFlow model using KenPom data that correctly predicted 2024 champion UConn, runner-up Purdue, and 11 of the Sweet 16 teams (69% accuracy for that round).

Code: github.com/415yoni/ANN_FinalProject