Aditya Dewan

Courses: Deep Learning (Ph.D.), Discrete Math, Computer Systems, Functional Programming, Math Finance

Clubs: Carnegie Mellon Racing (Built SVMs from scratch, 2x speedup for GPU midline generation), Quant Club (Goldman Sachs Quantathon)

Built Malloc (Dynamic Memory Allocator) and Fully Concurrent, thread-safe file system in C from scratch

• Implemented novel skip-connection-based Mamba model to remove multi-head-attention-based retrieval performance decline in SSM architectures
• Analyzed State Space Models and Mamba-2 bottlenecks on knowledge retrieval benchmarks (MMLU)

• Engineered React web-app to simulate textual Markov Chain models for 900+ professionals; used in Concord University study (demo)
• Designed efficient algorithm for automated graph generation and n-gram estimation for low latency on large text datasets

• Devised novel compression algorithm yielding 2-24% more accuracy in 10% training time; awarded by NSA, ISEF, WAICY
• Leveraged high-descent-potential saddle points via Hessian approximation; tested on MLPs, CNNs, ResNets
• Project: SPRKD

• Architected first ML infrastructure for min latency and high server volume; devised novel drug vector embeddings
• Deployed few-shot LLM drug monograph summarization API and React app for in-clinic patient diagnosis

• Led team of 3; developed + deployed recommendation engine API to 45k clients
• Engineered hybrid GAN, Gradient Boosted Tree, MLP architectures for production