Teaching

Graduate Teaching Assistant, University of Waterloo, Waterloo, Canada (January 2023 – August 2023)

Subject: CS 136 - Elementary Algorithm Design and Data Abstraction (Class strength: 1080) and CS 245 - Logic and Computation (Class strength: 200)

• Participate in the assessment process while collaborating with 33 ISAs, IAs, and TAs to provide students with effective, timely, and appropriate feedback to support their learning.

Graduate Teaching Assistant, University of Waterloo, Waterloo, Canada (January 2022 – April 2022)

Subject: ECE 657A - Data and Knowledge Modeling and Analysis (Graduate-level Data Science and Machine Learning course)

• Deliver a range of teaching and assessment activities including tutorials (in hybrid mode) directed towards delivering topics related to emerging tools and techniques in Machine Learning at a graduate level for about 150 students.
• Contribute to the development of appropriate teaching materials to ensure content and delivery methods meet learning objectives.
• Participate in the assessment process while collaborating with Prof. Mark Crowley and 2 other Teaching Assistants to provide effective, timely, and appropriate feedback to students to support their learning.

Industry

Artificial Intelligence R&D Engineer, Quadrical AI, Gurgaon, India (May 2019 – October 2020)

Framework: Python (Tensorflow/Keras), Flask, Postgresql, MongoDB, Docker, Microsoft Azure

• Created a real-time yield forecasting AI model using real-world unstructured sensor data for a leading solar company of India, which performed better than our business competitor, decreasing grid penalties by 90%.
• Developed a novel multi-objective supervised learning-based architecture to predict customer propensity for a leading e-commerce company in India. This model achieved a precision of 94%, beating their in-house model. (Real-world structured customer and product data)
• Developed a price prediction AI model for India’s No. 1 OTA (Online Travel Agency) with an ensemble of 5 novel architectures. (Real-world unstructured time-series data)
• Worked on preparing a standardized schema and developing the ETL (Extract, Transform, Load) pipeline for a leading solar company.
• Worked on backend development and DevOps to deliver the above products.
• Conducted interviews to hire candidates for the required engineering roles.

Research Internships

Summer Research Intern at Computer Vision Lab, IIT Madras, Chennai, India (May 2018 - July 2018)

Research Topic: Deep video captioning, Framework: Python (PyTorch) Supervisor: Prof. Anurag Mittal, Professor, Department of Computer Science and Engineering.

• A novel LSTM and CNN based architecture is proposed for deep video captioning using a combination of cross attention, action recognition, and object detection techniques.
• Increased the BLEU score by 40%, as compared to our previous vanilla model.
• Multiple captions for a particular video can be generated, depicting all the crucial instances.

Summer Research Intern at IIT Roorkee, Roorkee, India (May 2017 - July 2017)

Research Topic: Representation learning on text, images and acoustic data, Framework: Python (Keras/TensorFlow) Supervisor: Prof. R. Balasubramanian, Professor, Department of Computer Science and Engineering.

• A novel step-based correlation multi-modal CNN (CorrMCNN) model was proposed, which reconstructs one view of the data given the other while increasing the interaction between the representations at every intermediate step.
• Through extensive numerical experiments, it was found that the proposed model performs much better than the existing state-of-the-art techniques for joint common representation learning and transfer learning on images, text and acoustic data.

Summer Research Intern at SWAN Lab, IIT Kharagpur, Kharagpur, India (May 2016 - July 2016)

Research Topic: Quadcopter state prediction and wireless activation of sensor nodes, Framework: Python on Raspberry Pi Supervisor: Prof. Sudip Misra, Professor, Department of Computer Science and Engineering.

• An ANN classification model was used to predict different states of the quadcopter based on sensor data.
• A cost-effective mobile master-node architecture was developed with a burst power node activation, consuming more than 100 times fewer units of electricity in a week’s run as compared to the traditional architecture.