Sushmit Vaish

Teaching Assistant for Software Refactoring course

Working with the AGI team (Artificial General Intelligence) team in the field of AI/ML. Most of my work revolves around LLM model customization techniques. • Reduced LoRA inference GPU cost by 4× by pioneering a post-hoc adapter extraction technique preserving performance. • Optimized LoRA training time by 58% by minimizing computations using a novel fine-tuning approach. • Automated LoRA-enabled inference benchmarking using LLMPerf, generating key latency metrics (TTFT, TPOT).

Developed a "Next-generation auto grading: developing a multi-agent, LLM-driven code assessment platform". • Created an LLM-based auto-grading system with CodeLlama-7B, achieving 80% alignment with instructor feedback. • Applied vector embeddings, prompt engineering and advanced NLP metrics (ROUGE-L, BERTScore) for grading accuracy.

Teaching Assistant for the "Foundations of Software Engineering" course at CMU for grad students. • Mentored 20+ students in 'Foundations of Software Engineering' through Code Reviews, Recitations & Agile practices. • Facilitated backend project design and promoted modern SDLC methodologies in classroom and labs.

• Generated $270K revenue through process optimization initiatives. • Successfully led cross-functional teams of 2-3 associates, delivering multiple high-impact projects on time and within scope.

• Improved data extraction performance by 25% via Java multi-threading and Hadoop optimization. • Architected a data publish platform in AWS that cut internal effort by 70% and increased client refresh rates 5×.

• Automated quality checks in data processing pipeline to save daily runtime by 50% using bash scripts. • Maintained data processing pipeline for specific Pharma domains.

• Boosted app reliability by writing automated unit tests and embedding them in CI/CD workflows. • Resolved critical bugs and collaborated with developers in agile delivery cycles.

1. Foundations of Software Engineering - Developed a full stack web application using NodeJS and Express in the backend, MongoDB as Database and TWS (Vanilla web stack in frontend) 2. Functional Programming in Practice - Developed an Arbitrage Gainer project in F# for Bitcoins 3. How to Write Fast Code - Developed an efficient algorithm for Image Blurring techniques such as Averaging filters which was faster than the current version of Open CV implementation 4. Data Science in Software Engineering - I am pursuing this course to gain hands-on experience with Machine Learning and Data Analysis. 5. Software Design and Architecture - This course is about understanding and developing architecture for software systems based on the requirements. I am currently pursuing this course too. 6. Cloud Computing - This course gives an in-depth knowledge of how to use Microsoft Azure, GCP and AWS. 7. Research Project - Next Gen Auto Grader using LLMs: Multi-agent, LLM driven code assessment platform

I learned all my basic Computer Science fundamentals which includes Database Management, Operating Systems, Programming in C, C++, Python, Java, along with Data Structures and Algorithms and many more fundamental CS topics. This degree also included advanced topics such as Machine Learning, Artificial Intelligence and Internet of Things. I also completed a research project during this time. Details provided below - Data Extraction and Sentimental Analysis from “Twitter” using Web Scrapping: Used Machine Learning to train the model to extract data and identify the sentiments w.r.t posts & messages on Twitter. (Link - https://www.ijeat.org/portfolio-item/a2226109119/)