Aspiring AI/ML Engineer and Safety-Concious Developer
I'm an incoming Master of Engineering student at Cornell University, focusing on Artificial Intelligence and Machine Learning. I've graduated from the University of Maryland, College Park, with a Bachelor's degree in Computer Science, and a minor in both Statistics and Astronomy. I'm passionate about building projects that leverage AI/ML to solve real-world problems, and I'm particularly interested in the intersection of AI and safety.
A semantic search engine for ArXiv papers, built using Sentence Transformers, with searching capabilities enabled by FAISS and BM25. Implemented query expansion via Groq API to enhance search results.
Capstone for CMSC 435 -- Software Engineering at UMD. I worked in an Agile team of 5 developers to design, implement, and test a robotic arm simulation to allow UMD Nueromotor Control and Learning (NMCL) lab researchers to study Sense-of-Agency (SoA). We presented at the 2026 UMD Clark Engineering Capstone Design Expo
An Image Quantizer that relies on the K-Means++ algorithm.
My personal portfolio website, hosted on GitHub Pages.
I've worked under Professor Tom Goldstein and PhD candidate Monte Hoover on DynaGuard 2, a dynamic guardrails model for policy adherence in LLMs. I've collected over 220K+ AI safety and LLM interaction samples by analyzing major safety datasets such as CWEval, MT-SEC, ProSec, and RedCode, resulting in a comprehensive dataset for model training and evaluation. I then supported the model training phase by implementing GRPO and SFT in vERL and Modal. Additionally, I derisked future project phases by looking into existing red-teaming frameworks such as NVIDIA's garak and Microsoft's PyRIT, resulting in informed decisions on project direction.
Over two semesters, I assisted over 500 students in mastering complex data structures such as B-Trees, KD-Trees, and Blockchains. This involves time and space analysis, code implementation, and real-world applications of these data structures. I held weekly office hours to help students programatically develop structures such as Scapegoat Trees and Compressed Tries from the ground up. I proctored and graded 7 exams, maintaining communication with both students and the professor. See the course website here.
I worked on the Engineering team for a 501(c)(3) non-profit that builds software for other non-profits. As a member of the Operations team, I designed the backend database schemas and tables, clarified relations between tables, and implemented these designs through REST and GraphQL APIs using TypeScript. I wrote comprehensive tests to make sure all API functions worked as intended, ensuring 80%+ test coverage across the codebase.
I was on a developer team for a campus club dedicated to providing other clubs with websites and online platforms. I architected 10+ components in Figma, then later implemented them into dynamic, responsive webpages with React and Tailwind CSS. I collaborated in an Agile setting, managing tasks and contributing to timely project completion.
Interested in collaborating? Email me directly at desai.harsh.747@gmail.com.