I'm Sasi Kiran and I'm a Computer Science sophomore at Indian Institute of Technology Roorkee, interested in the science and implementation behind DeepLearning. The biggest driving factor for me is to better myself as a developer/ data scientist and also give back to the developer community which has taught me a lot.
I have worked on both Web and Android applications and am currently exploring the field of Data Science and Deep Learning. I am an experienced user of NoSQL databases, Object Oriented Programming and various Machine learning tools/ frameworks. I'm comfortable with web technologies like HTML, CSS, Bootstrap, JavaScript, jQuery, MongoDB and node.js etc. I'm fluent in various ML/ DL frameworks like NumPy, Pandas, matplotlib, scikit-learn, PyTorch, Tensorflow, OpenCV etc. I'm also familiar with web scraping using BeautifulSoup. The languages I use majorly for competitive programming and other tasks are C, C++, Java and Python. I've also used VHDL and MIPS for a few curriculum projects.
I've worked on Deep Learning projects like Neural Style Transfer using PyTorch, Name generator using LSTMs, Machine translation using Neural Attention Mechanism etc. and developed Android applications like PassOn - A collaborative cataloging app where users can upload/ request for used books and Anonytter - A twitter clone where anyone can post anonymously.
I've also developed an Android application that can classify user-taken image as containing any of the 10 skin diseases that we trained our model on. I used jQuery and BeautifulSoup to scrape over 5000 images from various online dermatology libraries and used OpenCV for image pre-processing. I trained the images on a ResNet-50 using transfer learning. The model reported a validation accuracy of around 80%.
I am driven to build applications to solve problems/ discomforts one might face and hopefully learn a lot along the way.
Projects
Automated Frontend Designer
Automated Frontend Designer is a python application that can turn a hand-drawn sketch into HTML code and instantly generate a webpage from the sketch at real-time.Deep Learning, OCR, Object DetectionCCTV Recap
Summarize hours of footage shot by static CCTV cameras into a short clip that shows all events as if they're occurring concurrently with timestamps as shown in the demo.NumPy, OpenCVSkills
Experience
- Google - GSoC Student DeveloperMay 2019 - August 2019
• Worked on identifying boundaries between different TV shows present in the same video and labelling the shows.
• Used Face Recognition, Image Clustering, Video Summarization, Scene boundary detection, Text analysis in the course of
the project.
• Drew statistical inference from Multi-modal, unstructured data and worked with a dataset of over 20,000 videos. - Machine Vision Lab, IITR - Research scholarFebruary 2019 - July 2019
Automatic Evaluation of Machine Synthesised Speech - Research project at Machine Vision Lab, IITR sponsored by Samsung R&D.
• Devised and coded an objective metric to automatically evaluate the humanness of Text-to-Speech systems
• Trained a Generative Adversarial Network on the LJSpeech dataset and used an ensemble model for evaluation - Srishti, IITR - DeveloperFebruary 2019 - March 2019
• Built a Python application that can turn a hand-drawn sketch of a webpage into HTML code and instantly generate it on
the browser in Real-time.
• Used TensorFlow object detection to detect and locate all HTML elements present in the sketch
• Used Tesseract OCR to read text to be filled in the HTML elements, rendered the output and displayed on the browser