This project involved designing and implementing an AI-powered communication assistant tailored for dynamic user interaction. I developed the overall architecture and handled core backend logic using Python and MongoDB. The assistant integrates custom intent detection and a Retrieval-Augmented Generation (RAG) pipeline built with Hugging Face models to interpret user queries and fetch relevant business data. Due to an NDA, source code and interface visuals are not publicly available.

For my final project in an AI fundamentals course, I built an image classifier using a convolutional neural network (CNN) in TensorFlow. The model was trained to distinguish between four Pokémon: Growlithe, Charmander, Dragonite (visually similar), and Greninja (an outlier). This setup was chosen intentionally to observe how the CNN handles intra-class similarities and outlier detection. The model was developed and trained in a Jupyter Notebook, and evaluation was done using matplotlib visualizations of accuracy, loss, and classification predictions. This project strengthened my understanding of convolution layers, overfitting, and data preprocessing.

This project was developed for my distributed systems class and focused on designing a robust backend architecture to unify heterogeneous sensor data from multiple greenhouses. Sensors from different brands sent temperature and humidity readings in various formats (Fahrenheit, Celsius, JSON, XML). I built a middleware service that authenticates senders, standardizes data into a unified format, checks threshold violations, triggers alert notifications via SMS, and persists data to MongoDB. The system includes microservices for sensor management, alarms (using Twilio), reporting, and a REST API for third-party analytics access. The middleware communicates asynchronously with microservices through RabbitMQ and interacts with the frontend via HTTP requests. Configuration data is stored in MySQL for relational consistency. Additionally, I developed a greenhouse simulator on an ESP32 device that connects to the middleware through WebSockets to mimic real sensor data. The backend services are primarily written with Nest.js and TypeScript, the alarm microservice with Python Flask, and the simulation in Arduino C++.

Developed as a team project for a web development class, CIA is a modernized clone of our university’s enrollment system. It provides two main interfaces: an administrator panel to configure courses and curricula, and a student portal for enrollment. The system enforces curricular planning rules by allowing students to enroll only in classes for which they have met prerequisites. It also checks for schedule conflicts to ensure a smooth registration process. The project used microfrontends for modular development and JWTs for secure authentication. Built with pure JavaScript, HTML, and CSS, CIA emphasizes simplicity and usability while handling complex enrollment logic.

This project is a basic 3D engine developed using Java and linear algebra concepts. It allows for hardcoded 3D models to be defined, rendered, rotated, and colored in a 3D space. I implemented the engine from scratch as a way to explore applied linear algebra and foundational graphics programming. The renderer handles perspective projection, rotation matrices, and real-time updates through a basic UI. This project helped me build an intuitive understanding of 3D transformations and low-level rendering logic.

This project involved developing the backend of a web application tailored for photography studios to efficiently manage their business information and appointment schedules. A key feature is the integration with a WhatsApp bot, allowing customers to easily inquire about available packages, book services, and schedule appointments through conversational messaging. The backend was implemented using Nest.js with TypeScript and MySQL for reliable data management. Although this project does not include AI components, it demonstrates strong full-stack backend development and API integration skills.