Feb 2021 – Apr 2022 · 1 yr 2 mos · New York, New York, United States

• Improve the company's verification platform using Machine Learning and other techniques to automate and intelligently default crowd/user tasks
• Lead various data gathering exercises to gather clean and labeled data for our classifiers and object detectors
• Experiment with state of the art Computer Vision architectures, such as training NFNets on our own data squeezing an additional 2% accuracy boost in the upper 90's range
• Develop and integrate a unified benchmarking platform to evaluate our Machine Learning models against standard benchmarking data sets

Jun 2020 – Dec 2020 · 6 mos · New York, New York, United States

• Experimented with few-shot learning methods for image recognition to tackle the cold-start problem in order to deliver pilots for document classification from easily available online form templates, removing the need for hand-labeled data
• Achieved accuracy boosts of up to 50% for small datasets using data augmentation & transfer learning (over and above an existing pre-trained VGG neural network on Imagenet)
• Implemented siamese network to predict labels by taking majority vote amongst an ensemble of VGG networks that compute the similarity between two data points through embeddings, gaining an additional 4% boost in accuracy

Sep 2019 – May 2020 · 8 mos · Greater New York City Area

• Tutored 100+ students on recent deep learning developments \& frameworks such as TensorFlow, Pyro, PyTorch; assisted the professor in modeling assignment questions, and mentored students for their term projects on state-of-art deep learning technologies like transfer learning, adversarial methods, 3D object detection, deep reinforcement learning

Sep 2019 – Dec 2019 · 3 mos · Greater New York City Area

• Worked with Prof. Hod Lipson towards conceptualizing and developing a real-time 3D Image Rendering system for building a Portable Ultra-Sound Machine
• Modeled and experimented with a novel technique for image enhancement using a combination of Machine Learning and Computational Geometry methods to filter out noise and echos from ultrasound images

Jul 2017 – Jul 2019 · 2 yrs · Bengaluru Area, India

• Led the research for conditional statement extraction and mapping for knowledge graph population from raw text which was used by a downstream conversational AI agent; achieving an accuracy of 90% against a baseline of 78%
• Deployed containerized micro-service for knowledge graph population to a kubernetes cluster part of a CI/CD pipeline
• Acted as a team focal for Retail and Distribution sync, working on evolving research strategy, project overviews, technical advances, open problems from across the lab relevant to Retail and Distribution
• Developed an active-learning based approach for precise entity extraction from IBM Redbooks; achieving the same accuracy of 90% by using just 35% of the original training data
• Designed and developed a ticket search tool for a client using Elasticsearch that included experimenting with data ingestion, query tuning and ranking results using ranker; Led to a 70% reduction in new ticket to agent mapping times
• As an active member of IROLICK, conducted month-end parties and other fun activities in IBM India Research Lab

May 2016 – Jul 2016 · 2 mos · Bengaluru Area, India

• Designed and statistically validated the in-process quality control and key product attributes valued by the consumers for biscuits
• Inspected the efficacy and sufficiency of existing quality checks for biscuits and suggested some improvements over the old ones and also explored various new technologies such as E-nose and E-tongue
• Conceived and suggested a roadmap achieving potential savings worth ₹3.6 million

Jul 2015 – May 2016 · 10 mos · Kharagpur Area, India

• Managed and conducted various events of the Department Society including fresher’s orientation, farewell, and some sports events

May 2015 – Jul 2015 · 2 mos · Mumbai Area, India

• Developed Phonetic Keyboard, a hardware-based transliteration solution, transliterating Roman characters into 9 Indian languages
• Took part in Wikipedia Hackathon and developed a tool for fetching sample text from Wikipedia and converting it into 9 Indian languages and uploading it back to the server
• Programmed an LCD based GUI (by programming ATmega328) for SPANDAC which is an Earthquake monitoring device

Aug 2013 – Feb 2014 · 6 mos · Kharagpur Area, India

• Worked extensively for publicity and management of Kshitij (the annual techno-management fest of IIT Kharagpur) and helped in conducting various workshops on manual and autonomous robotics.

Jul 2013 – Jan 2015 · 1 yr 6 mos · Kharagpur Area, India

• Designed the electronics (power management, sensors, data acquisition system) of K-2 (a student-built car for an international competition); analyzed sensor data using MATLAB for optimal calibration of sensors
• Developed a homemade optical wheel speed sensor using a simple LED-LDR circuit and a trigger wheel