Traffic violation detection

Developing an advanced traffic violation detection system using YOLO (You Only Look Once), OpenCV in python, to implement real-time detection of traffic violations, including traffic signal hopping and heavy vehicle monitoring. Making an auto fine deduction system using number plate detection with Tesseract, and tracking with DeepSORT. Utilized Python libraries such as imutils, PyTorch, and TensorFlow for machine learning and computer vision tasks. Collaborating with Mumbai Traffic Police to tailor the system for local traffic regulations and for testing purposes.

Google Photos classifier

Worked on an extensive project involving hand detection and gesture recognition, using different libraries in python. Implemented CNNs for face and hand detection using the TensorFlow, Keras, OpenCV, and Dlib HOG module. Developed a user-friendly interface for the project using the Tkinter, ensuring seamless interaction and functionality. Streamlined the deployment by converting the Python code to executables using PyInstaller, cx_Freeze, and py2exe.

Traffic Signal Optimization

Developed a mathematical model to minimize vehicle wait times, maximize throughput, and reduce emissions at signals. Utilized sequential quadratic programming to solve constrained nonlinear optimization problems with traffic flow. Ran tests with real-world parameters and visualized results to validate the efficiency of the model under constraints.

Bankruptcy in Poland

Explored data obtained from Polish economists studying bankruptcy, used JSON and terminal for compressed files. Made a bankruptcy prediction model using logistic regression, decision trees and tuned hyperparameters. Implemented techniques like random forests, gradient boosting with scikit-learn for model's robust performance.

Classifying building designs

Implemented K-means and hierarchical clustering to sort building designs into shape-based families, using Python. Conducted PCA for dimensionality reduction, ensuring efficiency with K value via Elbow and Silhouette methods. Enhanced model accuracy by detecting and analyzing outliers, improving cluster profiling and easy to interpret model. Trained a Random Forest classifier on labeled clusters, achieving performance metrics: precision, recall, and F1-score.

Air quality in Nairobi

Leveraged MongoDB queries to extract air quality data of Nairobi, Lagos, and Dar es Salaam from large databases. Applied time series data tools analysis to forecast PM 2.5 readings, using data preprocessing for predictive modeling. Developed autoregression models and optimized hyperparameters to predict air quality fluctuations for a day.

Housing in Buenos Aires

Developed a linear regression model in scikit-learn to predict apartment prices in the city of Buenos Aires, Argentina. Constructed data pipelines using simple imputer, encoding with OneHotEncoder, and linear regression predictor. Enhanced model performance by reducing overfitting and created a user friendly and intuitive dynamic dashboard.

Co-Creator | Sustainable Clothing Venture | Susclou

Started a venture dedicated to reducing textile waste and environmental pollution through fabric replacement. Successfully entered Level 1 IDEAS Cohort 11 program at Desai Sethi School of Entrepreneurship, IIT Bombay. Presented a pitch deck to a panel of judges, highlighting the current problem in the market and Susclou's solutions. Selected hemp as the optimal material through extensive research and built a sample website for selling using HTML. Received mentorship from IIT Bombay alumni and DSSE faculties, leveraging their expertise to improve our idea.

Sumo bot

Collaborated with a team of 3 to design and assemble a bot for a sumo competition, detecting and pushing opponents. Developed the control system, programming the Raspberry Pi Pico by utilizing the Micropython library in Thonny. Conducted tests to calibrate sensors and motors to work in the ring, using the time library for autonomous working. Developed strategies for effective attacking and searching, using four IR sensors, and two IR sensors for ring detection.

Automated density measurement of solids

Designed a machine that measures the density of solids using Archimedes' principle, and received a perfect AA grade. Designed a model using Fusion360 utilising motors and a platform to lower objects into water and a load cell sensor. Calibrated the load cell and coded the arduino UNO, lcd screen to perform the calculations, and to display results.