Selected Projects

We developed a neural network model for exoplanet classification system that automatically analyzes NASA TESS mission data to identify and categorize celestial objects as Confirmed Planets (CP), Planet Candidates (PC), or False Positives (FP). Our system addresses the challenge of processing the massive backlog of TESS observations that would take astronomers days to review manually.

Using a deep neural network trained on over 5,000 observations with 13 key astrophysical parameters, we achieve an accuracy of 78% or higher in classification. The system features a two-tier approach: first checking NASA's database using fuzzy matching (2% error margin) for known planets, then deploying our AI model for new discoveries. Users interact through an intuitive web interface where they can input individual observations or upload batch CSV files, receiving instant predictions with confidence scores, temperature analysis, and habitability assessments.

This project presents an educational SAP-1 processor implementation on a BASYS3 FPGA board that communicates with an Arduino Mega through SPI to provide real-time visualization of processor operations. The system runs a basic instruction set (LDA, ADD, SUB, OUT, SNG, HLT) using two separate finite state machines - one for handling user inputs and another for executing instructions.

Students interact directly with the processor using switches to select programs and input operand values, and buttons to advance through execution states. After each instruction completes, the processor sends its complete internal state as 40-bit SPI packets to the Arduino, which then displays this information on five seven-segment displays, LEDs, and through audio feedback. This setup allows students to actually see what's happening inside a processor as it runs, making computer architecture concepts much easier to understand without needing any external debugging tools.

Conducted market price prediction using machine learning algorithms, including ARIMA, SARIMA, SARIMAX, Prophet, and deep learning models with PyTorch. Analyzed price trends of selected goods based on multiple parameters and visualized results using Matplotlib to evaluate model performance and predictive accuracy.

Designed and implemented an ENIGMA machine on the Basys 3 FPGA using Verilog, simulating rotor-based encryption for alphanumeric input. Developed configurable rotors and real-time character encoding, with output displayed on 7-segment displays.

Designed and built a mini Sumo Robot using an Arduino and a Genesis controller card from a mini sumo robot kit. Equipped with dual motors, contrast sensors, and ultrasonic sensors to detect opponents and stay within arena boundaries. Programmed autonomous behavior for competitive performance in enclosed sumo arenas.

Designed and built a model smart road system featuring a functioning drawbridge, speed trap, and toll booth, integrated with multiple sensors and electronic components. Implemented communication between modules using MQTT over WebSocket for efficient, real-time control and monitoring.

Developed a crater detection program using OpenCV to accurately identify craters in planetary surface images. Utilized the Hough Circle Transform algorithm with enhanced circle merging to reduce false positives. Implemented features for crater counting and optional visualization through crater filling for improved interpretability. The main problem that my algorithm had was lighting differences caused by different light sources around the universe which makes detecting holes of planets/asteroids quite hard. Although it's not perfect, it gave me some idea and some experience about detection algorithms and image processing.

Developed an automated gate system using an HC-SR04 ultrasonic sensor to detect objects within 10 cm, triggering an SG90 servo motor to open the gate. Programmed an ESP8266 microcontroller for wireless control via the Blynk mobile app, enabling real-time distance monitoring through both the app and UART serial output.

I participated in a research project titled "Modelling Variability of Intensive Care Unit Demand: A Comparative Cost Analysis and Performance Evaluation of Fixed versus Variable Beds Arrangements with a Simulation Approach." The study addressed the sustainability challenges posed by limited ICU bed availability and high operational costs. Our goal was to compare the effectiveness of fixed and variable bed allocation strategies using simulation modeling.

I contributed to the development of an integrated simulation framework that incorporated a predictive model for ICU length of stay (LOS), accounting for day-to-day uncertainties in patient demand. The project aimed to inform healthcare policy by evaluating cost-efficiency and performance across different ICU management strategies. We placed third in the competition with this model and were invited to present our work at the ESICM LIVES 2023 conference in Milan, Italy. The study was also published in PLOS ONE and is available on PubMed Central.

Developed a gold price forecasting model using Long Short-Term Memory (LSTM) networks implemented in Keras. Preprocessed and normalized historical time series data, engineered input sequences, and trained the model to capture temporal dependencies for accurate multi-step price prediction.