Projects

Designed a state-of-the-art deep learning model for self-driving car trajectory prediction using LSTM with spatial and temporal attention, achieving significant error reduction from 80,000 (linear baseline) to 7.97 MSE on the Argoverse2 dataset.
Boosted model accuracy by over 30% through innovations like multi-head attention, temporal CNNs, and Gaussian trajectory smoothing — outperforming baseline LSTMs and CNNs by a wide margin.
Engineered input features from agent data, incorporating acceleration & social context to handle diverse urban driving scenarios.

Technologies: Python, PyTorch, Numpy, Kaggle

Pioneered a groundbreaking study on malicious tool-calling as a deterministic attach vector for LLMs, establishing new insights into potential vulnerabilities.
Designed and implemented adaptive prompts to trigger malicious tool-calling in LLaMa, achieving an average success rate exceeding 80% across 100+ tested prompts.

Technologies: Python, Ollama, HuggingFace, Google Colab

Lead a design team of 7 engineers for my Senior Capstone Design project, designing an IoT water consumption faucet attachment that measures and records isolated water flow, providing useful data metrics and recommendations to consumers via IOS app
Spearheaded the development of a webserver on an AWS LightSail instance, architecting using Nginx and Flask, and optimized to handle 100+ simultaneous IoT device connections with high reliability.
Innovated an IOS specific app using Swift and SwiftUI, incorporating a login procedure, JWT token verification to display user specific data and metrics.

Technologies: Python, NginX, Flask, AWS, PostgreSQL, Swift, MQTT

Developed a web-based developer toolbox using JavaScript, HTML, and CSS, offering tools like JSON formatting, Regex generation, and URL encoding/decoding to streamline backend development tasks.
Automated testing and deployment using GitHub Actions, with Playwright for unit testing achieving 96% code coverage, Linting, and HTML5 Validation to ensure code quality and cross-browser compatibility.
Designed a customizable, browser-based interface with drag-and-drop functionality, prioritizing usability and security by avoiding logins or user data storage.
Collaborated on prototyping using Figma and Miro, focusing on modular design and performance for backend developers.

Technologies: Javascript, HTML, CSS, Jest, Playwright, Github Actions

Prototyped and implemented a comprehensive pantry management system that tracks inventory, provides recipe recommendations, and integrates with a React application for user interaction.
Programmed robust backend services using Node.js and Express, handling CRUD operations for pantry items, user authentication, and third-party API integrations, with dynamic SKU # lookup.
Adopted OpenAI API to generate personalized recipe suggestions based on current pantry contents and user-defined parameters.

Technologies: Node.js, Express, MongoDB, RESTful APIs, OpenAI API, AWS