Hello, I'm Karthik from Earth Dimension C-137. I'm a graduate student at Carnegie Mellon University, specializing in Machine Learning and Natural Language Processing.
I have multiple publications to my name including one on neural program execution which was published at ICLR 2017. I'm also a hackathon enthusiast and have won multiple hackathons at national and international levels.
I love teaching. In my free time, I take training sessions for freshers and mentor them at endurance. The idea of imparting my knowledge to enrich someone else's holds a special place in my heart.
You contact me at email@example.com or at +1-724-506-0009
Researched deep-learning based program induction and execution models.
Conceptualized “Program Embeddings” – a vectorized representation of Assembly Language program statements (eg. ARM, MIPS).
Augmented the Neural Turing machine with novel ways to access large main memory, a fuzzy register bank and an instruction bank.
Ensembled Neural Networks whose execution was governed by the NTM controller and program counter to learn to execute ARM code from examples.
Published the outcomes of my research at the International Conference for Learning Representations, (ICLR Workshop Track) Toulon 2017. Open Sourced the code, dataset used for training and demo at neu0.github.io
Jarvis – Conversational Form Chatbot which runs
diagnostic tests against a website and offers resolutions.
Models different tests as a tree and identifies the most important
issues by asking users questions to aid its inference.
Implemented a tf-idf based related tickets finder to reduce SLA for
Bigrock Instant - Architected a RESTful microservice based order management and support system for building instant websites. Built a central authentication system using JWTs and Redis behind a Netflix Zuul Gateway.
ESCache Search - Developed a CQRS based Search service catering to search-as-you-type requests using ElasticSearch on SpringBoot.
Orderbox - Maintained the retail division of order management platform which powers 3.2M customers. Added API authentication for customers & integrated new services like Office365 and payment gateways.
Automated form onboarding process for
Improved overall efficiency by 80%
Inspired by Neural Style Transfer which transfers an image’s style to another image; Sought to transfer writing style while preserving semantic content
Investigated the feasibility of casting this as a Machine-Translation problem
Scraped a Shakespearean-Modern English parallel dataset from Spark Notes
Implemented a deep Encoder-Decoder with attention in TensorFlow.
Designed a generic Neural-Network framework based on the computational graph concept using NumPy and first principles to complement my theoretical understanding.
Implemented Feedforward, Recurrent and Convolutional Neural Networks.
Implemented AdaGrad, RMSProp; Cross-Entropy and Mean Squared loss.
Benchmarked the implementation with MNIST, CIFAR 10, and PTB datasets.
Compared RNN, CNN and MaxEnt models for Twitter text-classification.
Conceptualized a system capable of monitoring supermarket shelves and automatically reacting to events such as mislabeled or out-of-stock products.
Implemented image segmentation using the Watershed algorithm using OpenCV to yield (product, label) tuples for a given image of the supermarket shelf.
Built a product detector using image feature extraction and a KNN classifier. A CNN was not used due to sparsity of training examples.
Built a Seq2Seq OCR engine using a Convolutional Stack Encoder and Recurrent Stack Decoder with CTC log-loss objective function. Trained first on synthetic data (due to sparse training data) followed by transfer learning.
Presented the work at ABinBev HackTheWorld and placed 3rd out of hundreds of contestants comprising both students and working professionals.
Designed a fault tolerant master-
slave based distributed data store
Architected an event driven frame- work to ensure consistency and high availability of data
Implemented orchestration among nodes for data replication
Designed a hybrid of LSTM and MaxEnt models to perform phrase level sentiment analysis on twitter data
to understand how Indian banks fared with demonetization.
Extracted phrases with Stanford NLP Parser and visualized the results of the hybrid network.