"Be inspired to make a difference one code at a time."
Hello!
I am pursuing Masters in Computer Science at Columbia University. I completed my Bachelor's from Birla Institute of Technology and Science, Pilani in Computer Science Engineering
I am looking for opportunities to work in applications at the intersection of Machine Learning, Deep Learning and Natural Language Processing. I spent this summer interning with Search Team of SAP Ariba, Palo Alto. I worked on building deep learning models to improve the search relevancy. My undergraduate dissertation was in the topic of Image Captioning under Prof. Vinay Namboodiri and Prof. Sujala Shetty.
When I'm not working, I am usually travelling and exploring new places. I also love to paint, sketch and solve puzzles in free time. I love meeting new people and learning new things, so please feel free to say hello and share a story with me.
You can find my resume here.
Past Affiliations
 |
 |
 |
 |
 |
Research Interests & Projects
During my undergrauate degree, I tried working on multiple things to discover my interests. It was in my junior year when I cam across Machine Learning. For me, it is the ability to identify complex patterns in millions of data. What really happens is that the machine learns through an algorithm that checks the data and is able to predict future behavior. And automatically, implies that these systems improve themselves over time, without human intervention. It soon struck me ML can solve real life problems across such an elaborate number of fields. My further involvement in research has given me first-hand exposure to the process of active scientific research, resulted in incredible research experiences and career opportunities, and instilled in me a passion for further exploration.
Interests
- Machine Learning
- Natural Language Processing
- Deep Learning
Projects
-
The idea was to generate romantic stories associated with any input image. First, a image captioning model was developed to generate captions that captures the various aspects of the images verbally. Parallely, a RNN encoder decoder was designed using BookCorpus dataset which maps each passage to skip thought vector. Passing the captions (preserve thought) produced through RNN encoder decoder to produce romantic stories for the image whose captions were used.[Report] [Slides]
-
A web application was developed which enables sharing of images with users and downloadable in different formats and sizes. The backend of the app was designed in Python (Flask) using MySQL as our database and front end designed in React. Dev, stagging and production separately maintained on GCP and were updated using CI/CD (Circle CI). This was a project implemented as part of the course for Advanced Software Engineering using Agile methodologies.[Code]
-
Manual annotation is an extremely painful process so a NER API was developed for History Lab to tag entities from raw data from SQL server. Varaition of Matching algorithms to verify persons names in the NER tagged data from wiki or from other sources. Based on the retrieved entities, Social networks were made using NetworkX, Gephi and Stata to unerstand the relationship between the entities obtained.
-
A ML model was built a model to predict the unified Parkinson’s disease rating scale of patients from the audio features of noninvasive speech test done through telemonitoring using Python programming language. Designed an algorithm such that its accuracy was found to be comparatively better than conventional regression models. Paper based on the project, "Improving Accuracy in Noninvasive Telemonitoring of Progression of Parkinson’s Disease using Two-Step Predictive Model" was presented in EECEA, 2016. Later, it was published in IEEEXplore and also Research Gate. [Published Paper][Raw]
-
I worked with the UAV Team for two years. We various mini projects in this time frame. For indoor navigation in GPS denied environments, we implemented SLAM (Hector & Gmapping) using LIDAR Sensor. Also for outdoor we implemented Way-Point Navigation [Video]. Worked with Emerging Technologies - Leap Motion, Fat Shark, 3D Printer.
International Aerial Robotics CompetitionTo solve the 7th Mission Problem statement we implemented navigation using just cameras and LiDAR and numerous other algorithms like Optical Flow, Visual Position Estimation, Canny Edge Detection. Details present in [Report]
Applications built - Developed 3D Mapping Application for Star Cements; Worked with lnfini to develop a Hybrid VTOL. Received “Drones for Good Award 2016” with our submission “Smart Inspection of Solar Panels”.
Work Experience
-
In order to improve the relevancy of search results, deep learning model was designed to predict the category of product from search query to enrich the query. A Branched CNN was implemented with Conceptnet Numberbatch and ELMO embeddings (Keras/Tensorflow) to give an accuracy of 87%. A web application in Flask was built that takes about 70ms per inference to show predictions. Ultimately multiple versions of the model were deployed in production environment using Tensorflow serving and gRPC (20ms/response).[Report]
-
Replicated the torch implementation results of Dense Captioning model to obtain region-specific captions. Refined the captions by putting filter on the overlap area and caption similarity to reduce redundancy. Categorized captions through topic based clustering and fed them into an encoder-decoder model trained to generate sentences from phrases obtaining successful results in producing paragraphs for images in Visual Genome Dataset.[Report] [Slides]
-
Web Development using HTML5,CSS, Javascript was done to revamp their existing applications. Automated the generation of their Inventory Report made daily using Visual Basics for Applications. Improved the Middle East Dashboard, that was a detailed dashboard that described sales of individual products and country-wise sales of each product.
