My name is Samanyu Satheesh, and I'm a grad student at UC San Diego. Thanks for taking the time to visit my page! Here's a little about me:
Born and raised in Mountain View, California, I grew up in a diverse community surrounded by the cultures of the South Bay. I completed my BS at Cal Poly Pomona, majoring in Computer Science, where I graduated Summa Cum Laude in 2024. I then joined eGain as a Solutions Engineer, where I refined and built upon their core SaaS offerings, developing AI-driven projects that increased engineer productivity. Following this, I enrolled as a Master's student at UC San Diego, majoring in CS with an emphasis in AI and Machine Learning.
In my free time, I love to play the drums, listen to music, work out, watch football, and eat food. I enjoy spending time with friends and family, as well as learning new things. I'm currently learning more about machine learning and its applications in healthcare.
You can scroll down to see some of my recent work!
I built a Retrieval-Augmented Generation pipeline using LangChain to answer questions over technical documentation, systematically tuning chunk size, embedding model, search type, and reranker settings across 30+ configurations. The final system used a GPT-OSS-120B generator, FAISS vector store, and a cross-encoder reranker.
I fine-tuned small language models (≤2B parameters) using LoRA adapters to perform schema linking, identifying which tables and columns a natural language question references as a core step in NL-to-SQL systems. Testing 30+ configurations across Qwen, Llama, and SmolLM2 model families with NLTK-based schema filtering and hallucination post-processing, the best result was achieved with Qwen3-1.7B using attention-only LoRA.
To investigate the applications of Machine Learning in biomedical imaging, I built neural networks that reconstruct high-quality MRI images from just 25% of the usual scan data, potentially cutting patient scan times from 20 minutes to 5. I compared two approaches: one leveraging raw frequency-domain physics, the other learning to remove artifacts directly, achieving a structural similarity index of 98%.
During my internship at PM Accelerator, I collaborated on a full-stack college recommendation platform that helps users find schools based on their academics and interests. We designed the React frontend for a smooth experience and contributed to the AWS-hosted backend with custom APIs, real-time search, and compatibility scoring.
I developed a web application that generates a recipe for any meal as well as a summary of the dish, ingredients, serving sizes, and step-by-step instructions. I used Flask for the backend and leveraged the OpenAI API for integrated prompting. This was a great introduction to applied AI and prompt engineering.
I collaborated on the Lockheed-Martin Fire Detection Project at Cal Poly Pomona, leading the ML team to train a model capable of identifying fires and smoke. We trained a YOLOv8 model using a few thousand images to achieve an accuracy rate of 80%, and I presented our findings to Lockheed-Martin engineers at the College of Engineering Research Symposium.
I worked on a research project utilizing the Boston Dynamics SPOT robot to perform autonomous package deliveries. SPOT was programmed to identify a package, retrieve it, and navigate its way to a delivery destination using fiducial markers. We presented our project at the College of Science Research Symposium as well as at an Open House showcase for new students.
At the 2024 Cal Poly Pomona BroncoHacks, I collaborated with a team to develop a full-stack campus-exclusive marketplace in under 24 hours. Built with Next.js, React, TypeScript, and Material UI, the platform supported user authentication, item listings, and filtered search, with Firebase powering real-time cloud data storage. This experience strengthened my skills in web development, teamwork, and rapid problem-solving.
In my senior year, I studied the NVIDIA CUDA programming language, learning how to harness the massive parallelism of GPU architecture to accelerate complex computations. Developed in CUDA and tested on systems at the National Center for Supercomputing Applications (NCSA), I learned to optimize performance for real-world data-heavy tasks like Convolution and Pooling, techniques used in Machine Learning.
If you'd like to get in touch, feel free to reach out via email or connect with me on LinkedIn and GitHub using the links below: