Hello!

During my Data Analytics Internship, I developed an Excel/VBA macro system for a company that automated the full invoice workflow by transforming cleaned financial data into standardized outputs and generating hundreds of customer invoices each month with built-in validation and tracking. To power the process upstream, I built a Python-based PDF parsing tool that extracted structured data from overseas shipping invoices, automatically collecting key fields and converting them into machine-readable datasets. The extracted data was then fed directly into the macro for aggregation, invoice generation, and tracking. This end-to-end automation pipeline significantly reduced manual data entry, minimized errors, and improved efficiency during monthly reporting cycles.

This research project explores using transformer-based language models to translate natural language queries into structured SQL statements. Leveraging fine-tuned sequence-to-sequence architectures, I evaluated performance on benchmark datasets and implemented constrained decoding techniques to improve query validity. The project emphasizes model evaluation, experimentation with hyperparameters, and understanding the limitations of large language models in structured query generation. It represents a blend of machine learning theory and practical implementation.

I designed and implemented a fully normalized relational database using SQL. The project involved requirements analysis, entity-relationship (ER) modeling, schema design, and implementation in a relational database management system. I defined primary and foreign key constraints to enforce referential integrity, normalized tables to reduce redundancy (up to 3NF), and developed complex SQL queries involving joins, aggregation functions, subqueries, and grouped reporting. The system supported structured data storage, retrieval, and analysis while demonstrating strong understanding of relational design principles and query optimization fundamentals.

As part of an undergraduate research initiative, I contributed to the development of a modular underwater drone control system. The platform utilizes an Arduino Mega to coordinate thruster control via ESCs using I2C communication and PWM signaling. My focus has been on reliable motor control, communication stability, and hardware integration in a constrained aquatic environment. This project combines embedded systems, power electronics, and systems integration in a research-driven setting.

Designed and deployed a real-time facial detection system on the Arduino Nano 33 BLE Sense using TinyML techniques. Trained and optimized a lightweight convolutional neural network (CNN) for binary face vs. non-face classification, then quantized and converted the model to TensorFlow Lite for Microcontrollers (TFLM) for on-device inference. The system processed camera input locally without cloud connectivity, enabling low-latency, privacy-preserving edge intelligence on a resource-constrained microcontroller. Significant effort was placed on model compression, memory footprint optimization, and inference speed to fit within the Nano 33’s RAM and flash limitations. This project strengthened my skills in embedded AI deployment, model quantization, memory profiling, and real-time edge inference.

This embedded systems project involved designing and building an automated pet door controlled by motion sensors and servo motors. Using a PSoC microcontroller, I integrated PIR sensors for detection and implemented PWM control for dual servo actuation. The system was designed to detect directional motion and actuate a mechanical drawbridge-style door accordingly. This project required hardware integration, signal conditioning, firmware development, and iterative physical prototyping, providing a comprehensive hands-on embedded systems experience.

This project involved designing and implementing a 32-bit structural RISC-V processor in VHDL. The architecture was built from the ground up, including the ALU, control unit, register file, data memory, and program counter, all connected structurally to form a complete single-cycle datapath. I developed and simulated the processor using a custom testbench to verify correct instruction execution and register updates. The goal was to deepen my understanding of computer architecture and hardware abstraction by translating textbook processor theory into a working, simulated system.

I modified and extended the Embedded XINU operating system at the kernel level. My work included implementing system calls, process creation logic, stack management, and UART driver functionality. By working directly in C within the kernel environment, I gained hands-on experience with process scheduling, memory allocation, and low-level hardware interaction. This project strengthened my understanding of operating systems by bridging theory with real system-level implementation and debugging.