Hardik Nahata Staff Machine Learning Engineer Large Language Models Personalization LLM Agents Generative AI Deep Learning Natural Language Processing Machine Learning ML System Design Software Engineering

NYC 311 Complaint Type Prediction

· 311 is a service that New York City residents can use to make non-emergency reports. The NYC 311 Dataset was made public by NYC OpenData and has about 40 columns and around 21 Million Rows.

· I have experimented with Traditional ML Models and Transformers to Predict the Type of the Complaint with the complaint text leveraging Natural Language Processing.

· Please use the link below to access the Jupyter Notebooks and Code.

Movie Recommender System

· Based on a movie dataset with large number of entries, this system predicts the best recommendations for a user through his/her past ratings.

· Used Item-Based Collaborative filtering and found the correlation values between every two movie ratings to predict the recommendations.

· Used Pandas library for operating on the .csv files.

Drowsiness Detection during Commute

· Created a prototype for a drowsiness detection system using a Webcam.

· An alarm is sounded when the driver falls asleep.

· Used the concept of Eye Aspect Ratio introduced by Soukupová and Čech in their 2016 paper, "Real-Time Eye Blink Detection Using Facial Landmarks".

License Plate Number Detection

· Performed various image processing algorithms like Dilation, Erosion, Edge Processing to preprocess the image

· Smoothened the Vertical Histogram through Low Pass Filtering

· Found probable segments for License Plate and performed Region of Interest Extraction.

· The code was developed in MATLAB.

Handwritten Digit Recognition using CNN

· The MNIST dataset contains a total of 70,000 handwritten digit images to train and test.

· Trained the data through a set of layers including Convolutional layer, Rectified linear unit, MaxPool layer and Fully Connected layer.

· Achieved an accuracy of 97% on the Test set.

Smart Parking System

· The Ultrasonic Sensor[HC-SR04] is placed on the parking spot which determines the presence a car.

· The signal is indicated through a High power LED placed at a certain height or on to the ceiling.

· The setup consists of an Arduino board, sensor and LED. Hence its highly cost effective.