I'm a Software Developer with a keen interest in Machine Learning. My experience includes working on Computer Vision problems and developing Geospatial web applications. I always try to gain an in-depth understanding of the technologies and concepts I use in my work. I'm excited about learning new technologies and facing interesting challenges. My coding skills include JavaScript, Python, and its ML libraries like scikit-learn, Pandas, and Pytorch. I am actively seeking Full Time Opportunities as a Software Developer/Machine Learning Engineer.
-
Experience
Web Developer (May 2018 - Present)
• Developed a geospatial web-app that helps maintain the Georgia Tech campus tree inventory of 15000 trees.
• The front-end is built using the ArcGIS API for JavaScript, jQuery and its plugins, and the backend is supported by ESRI’s ArcGIS Server.
• Enhanced the UX by adding querying and filtering capabilities and dynamically updating the map to reflect these operations.
• Integrated data addition and updation into the app saving hours of time spent in manually updating the database from an excel file.
• Decreased future app build times by modularizing the code into specific functionalities and reusing them.
Machine Learning Researcher (Jan 2019 - May 2019)
• Developed detection and tracking algorithms to track mosquitoes in videos to help study the spread of mosquito-borne diseases.
• 40+ hours of mosquito videos shot in a lab is background-subtracted into natural indoor environments to generate training data.
• Used 3D-CNN to capture temporal information and mitigate the low signal to noise ratio. Previous frames help localize mosquitoes in current frames.
• Achieved 95% detection accuracy for indoor environments.
-
Projects
1. Book Recommendation System
• Built a book recommendation system by scraping data for 140k users, 450k books and 2M friendship connections from Goodreads
• Used hybrid algorithms to boost performance(Precision0.45, Recall0.6)
• Designed a web-app using Flask and D3.js for users to graphically explore friends and books network
2. Neural Network based Soccer Game Model
• Built a deep neural network to model soccer games that could predict possible future states of play from past states
• LSTM architecture models the game as a sequence of states(player-ball positions) which lead to a distribution of future states. Special attention is given to player-ball interactions by training a fully connected layer
• The model learns salient features of the game and gives reasonable predictions for time-stamps closer to the input
3. Simulating NYC Citi Bike sharing platform
• Simulated the NYC Citi Bike sharing platform with 700 stations and
12000 bikes to study the imbalance problem wherein too many or too few bikes get accumulated at few stations
• Using greedy heuristics to optimize the initial distribution of bikes yields 40% more revenue than a random or uniform distribution