Distributed Data Parallel ML Training | Big Data Systems Course Assignment

  • Deployed, compared and contrasted different Distributed ML Training frameworks such as gather-and-scatter, allReduce and Pytorch DDP (Distributed Data Parallel) on a multi-node cluster setting.
  • Achieved 1.4 - 2.8 times speedup over a single machine using above techniques for gradient synchronization. [Github]

BadgerDB | Database Management Systems Course Assignment

  • Implemented Buffer Pool Management in BadgerDB using Clock Algorithm and Buffer Hash Table.
  • Developed a B+ Tree Index Manager in BadgerDB to improve query searches and range scans. [Github]

Cloud Based Real-Time Big Data Analysis | Big Data Lab Project

  • Implemented DataProc Cluster based Spark Job Computation for Data Processing & Model Training.
  • Computed Real-Time Predictions on Test Data by Spark-Streaming the Data to Apache Kafka Cluster. [Report][Code]

American Express - Cricket Analytics based on Deep Learning | AmEx Ignite Challenge 2019

  • Developed an LSTM-based Model for Real-Time Win Probability Prediction in the Game of Cricket.
  • Predicted 82.7% of the Match Outcomes Accurately at the end of the 10th over (out of 50 overs). [Paper]

Accident Casualties & Severities Analysis | Data Analytics Lab Project

  • Investigated Feature Importance using supervised ML techniques: Lasso, LGBM and Random Forest.
  • Categorized UK Districts into different Clusters based on Safety Levels using unsupervised ML algorithms. [Presentation][Code]

MovieLens Ratings Prediction | Machine Learning Course Project

  • Competed in a Kaggle contest (ranked 3rd of 60) to predict user-based movie ratings using machine learning.
  • Implemented collaborative filtering methods such as matrix factorization, regression & neighborhood model. [Report][Code]

Video Captioning using ActionVLAD and Bidirectional LSTMs | Deep Learning Term Project

  • Implemented “ActionVLAD” for Feature Aggregation which functions similar to the Attention Mechanisms.
  • Developed Bidirectional-LSTM based RNN decoders with Word2Vec Embeddings; evaluated on BLEU Score. [Report][Code]

Shape Matching & Retrieval | Digital Video Processing Course Project

  • Compared 2 conventional Shape Matching CV algorithms: Shape Context Descriptors vs Chamfer Distance.
  • Implemented Hungarian Bipartite Matching algorithm & reported Precision@20 score on MPEG-7 dataset [Report]

Credit-Card Defaulter Identification | Introduction to Data Analytics Course Project

  • Classified the Applicants who will Default in next 12 Months using their Credit Card Application Information.
  • Performed MICE, Rank & Plot for Data Imputation, SMOTE for Class Imbalance & LGBM for Classification. [Report]

Priority Queue, 4-nary Tree & Knapsack Problem | Data Structure & Algorithms Project

  • Designed a Network Scheduler to manage a Sequence of Network Packets using FIFO & Priority Queues.
  • Simulated Fluid Particles’ Interaction by mapping the Coordinates to a 4-nary Tree using the Binary Digits.
  • Solved the Expanding-Knapsack Problem using Dynamic Programming based Recursive Algorithm. [Report][Code]

SpaceX Hyperloop Pod Competition 2019 | Host: SpaceX, California

  • Avishkaar Hyperloop of IIT-M was among the 22 Finalists worldwide to Race on the SpaceX Test Track.
  • Contributed towards Ideation & Design of Subsystems, especially the Control Unit & Braking Mechanism. [Link]

Ad Hoc Wireless Communication Network | Inter IIT Tech Meet 2018 Project

  • Secured 2nd Position in the Inter IIT Tech Meet 2018, one of the Largest Tech Meets in India.
  • Developed a Hand Gesture Recognition Module using CNNs & SVMs for a Wireless Communication System. [Code]

Self-Balancing Robot | iBot Club Project, Center for Innovation

  • Innovated a Two Wheeled Robot which Adaptively Balances Itself, similar to a Segway.
  • Implemented PID Controller using Inertial Sensors to control the Motor. [Presentation][Code]