Harsh
LinkedIn:https://www.linkedin.com/in/harsh-ai/
Kaggle:https://www.linkedin.com/in/harsh-ai/
Email: harshris2314@gmail.com
A passionate and ingenious developer with prospenity to compete in fast paced enviornment and revamp readily to new challenges. A machine Learning enthusiast who is ready to work in the field of machine learning and Artificial Intelligence
Education
- B.Tech Guru Gobind Singh Indraprastha University(ECE)
My Skills
MooC Undertaken
- Deeplearning.ai Specialization
- Udacity Secure and Private AI Nanodegree
- Machine Learning In Finance Specialization
- DataCamp Data Analyst Track In progress
- DataCamp DataScientist Track In progress
Recent Projects
Skills Used - Python, PyTorch, Torchvision
The main task was to detect the Brand Logo. I used Flickrlogos-32 dataset containing logos of 32 brands like Adidas, Apple etc. I used the torch and torchvision libraries. I created a LeNet-5-liked network for the Brand Logo classification. Some of the major steps undertook were Edge Detection, Morphological Processing and Sampling method. YOLOV2 also serves well in this task.
Skills Used: Python, Numpy, OpenCV, Dlib
I made a facial Recognition system by implementing Siamese Neural Networks, the main task was to sucessfully identify the facial landmarks of a given face in real time,Inception network was hardcoded for this purpose without the use of any modern and scalable deep learning library.Dlib Facial Recognizer was used to find the facial landmarks and opencv was used for preprocessing and handling the image data
Skills Used: Tesseract, Opencv, Python, TensorFlow
This project is an OCR for identifying tables from Images and extracting data out of it, currently in development phase has the capability of identifying tables correctly using Computer Vision and extracting text data from tables to csv for future uses, A Deep DeSRT model architechture is being implemented to automate the task further and resolve redundancy,in tuning issues SK
Skills: TensorFlow,keras,Python,Plotly
The project used a fine tune Arima model with Wavenet architechture for stock market prediction of a single task based on various factors like previous price over the years, sentiments of people, evaluation of company's holdings.