Student · Developer · Researcher
Abhirama Sonny is a student and developer focused on Computer Science, Robotics, and Molecular Biology, building technical solutions to solve problems.
Student, programmer, and avid researcher in the fields of computer science, molecular biology and robotics.
Machine Learning, Computational Biology, Robotics, Software Engineering, Data Science, Quantitative Finance, Cybersecurity
Exploring technology and research through innovative solutions and projects.
Built an application that uses a Convolutional Vision Transformers (CvT) to extract key metrics from echocardiogram videos. It then uses a Retrieval-Augmented Generation (RAG) pipeline with vector search to feed DeepSeek r1 (an LLM) sufficient context for an accurate cardiac diagnosis, achieving 93% accuracy in diagnosing the condition. This project is aimed to be used by Cardiac Sonographers as a tool to cross-check their values, and by patients to understand their condition better.
Co-authored a JISEM paper proposing a peer-to-peer framework that dynamically routes AVs by sharing real-time position, velocity, and environmental data to preempt collisions and reduce congestion. In this system, a car sends the aforementioned data to its neighbors, who then use this information to optimize their own routes. This then allows an algorithm to calculate the optimal path for each car, taking into account the data from its neighbors to reduce traffic and collisions.
Developed a custom Java-based neural network framework, from scratch, supporting custom datasets via flexible CSV loaders, dynamic layer configuration, and multithreaded BLAS-style linear algebra routines. Integrated many activation functions (ReLU, LeakyReLU, PReLU, Sigmoid, Tanh, etc), dropout layers, L2 regularization, and grid-search hyperparameter tuning (learning rate, momentum, batch size). Validated on the MNIST dataset, achieving over 98% accuracy within 15 epochs and showcasing rapid convergence and scalability.
For each of the last three years, there have been over 600 mass shootings, almost two per day on average (gunviolencearchive.org). After the Uvalde and the Allen Premium Outlets tragedies, a faster law-enforcement response could save many lives. This project uses preexisting CCTV infrastructure to run a convolutional neural network (mean Average Precision > 90%) to detect a person carrying a gun in live video streams. Upon detection, it immediately sends authorities a snapshot, description, and precise location. Training data comes from the Internet Movie Firearm Database (imfdb.org) and a custom annotated image repository.
I designed a few-shot learning framework for diagnosing cantu syndrome, a rare genetic disorder, using a multi-modal approach that combines clinical data, genetic information, and medical imaging. The framework utilizes a transformer architecture to learn from limited labeled data, achieving high accuracy in identifying cantu syndrome cases. This project addresses the challenges of diagnosing ultra-rare diseases, using an attention mechanism with 2 layers of training to overcome low sample sizes.
I created a simplified programming language called Jaithon, inspired by Python and Java, to introduce elementary school students to the fundamentals of programming in an engaging and accessible way. Jaithon includes features such as bootstrapping, which allows the language to compile itself, and an efficient garbage collection system to manage memory automatically. Additionally, I implemented a custom parser/lexer system to process and interpret the language's syntax, enabling students to write and execute code with ease.
Diverse experiences shaping my growth as a student, programmer, and leader.
FIRST is a global nonprofit organization that helps young people learn STEM skills through robotics. I’m a programmer on FTC Team 7172 Technical Difficulties.
Co-founded a private VEX team as a programmer.
The Cambridge Centre for International Research (CCIR) provides research opportunities for high school students, by connecting them with postdoc researchers.
Led and coached multiple VEX teams across AISD to world championships.
Inclusive Computing Initiative teaches CS to children with special needs.
Vipravrinda is a nonprofit organization uniting the Trimathasta Kannada Brahmin community in America. I developed the website vipravrinda.org for the organization.
Recognition and achievements in competitive academic fields.
I scored 39/50 on the USA Biology Olympiad Open Exam, placing 4th out of 5,095 nationally. I achieved an Honorable Mention on the USABO Semifinal examination with a score of 78.4.
Promoted to the Gold Division in USA Computing Olympiad during the February 2025 competition.
I won 2nd Place at DRSEF, and was a finalist at the Texas Science & Engineering Fair for my work on Deep Learning Powered Classification and Analysis of Standard Heart Views in Ultrasound Imaging for Enhanced and Quicker Cardiac Diagnosis.
My PicoCTF placed in the 4% percentile of all participants. I contrubuted 1535 points out of 3710 points to my team.
I was awarded 2nd Place at the Great Scientist Soiree by UNT and Wipro for the HealthBuddy ML sensor device that monitors heartrate and blood oxygen via a necklace pendent.