Portfolio

---

Work Experience

Zeotap India Pvt. Ltd.
Data Scientist Intern
•Worked on cleaning, scoring, and graph embedding in huge network graphs(ID graph) with nodes and edges in Billions using state-of-the-artscalable algorithms in pyspark using graphframes, and networkx libraries.
•Completely designed and implemented the data partner(dp) scoring pipeline which scores a data partner depending on the anomalous data given by that data partner. Data Partners are scored based on 8 metadata functional features of the graph.
•The data partners(dps) are also scored using comparative feature vector-based scoring wherein each dp is assigned a feature vector based onstructural features and a few metadata features of the graph and an algorithm inspired from PageRank is used to score the data partners.
•Cleaning pipeline - Found connected components within the graph and applied clustering algorithms like Power Iteration Clustering and Lou-vain Detection to find communities within large connected components. Studied the graph embedding algorithms like GraphSage and GraphAttention networks(GAT) to find the Low-dimensional vector embeddings of nodes in large graphs.

---

National Taipei University of Technology
Intern
• Worked on a Large Parking Station Management System for the China Motor Corporation(CMC) under professor Leehter Yao.
• Wrote a program to efficiently schedule the charging interval of a vehicle in the parking lot based on its departure time and current battery charge status of the vehicle. Linear Programming with the GLPK library was used to perform this task.
• Designed a website application that shows the status of the Parking lot(Number of Vehicles, their current Battery Charge etc).
• Worked on CANBUS to enable the communication between the Battery management system and the charge management system.

---

Monash University
Research Intern
• Worked on a Research Project in Virtual Reality(VR) under Dr Jarrod Knibbe to predict pointing behaviour during room-scale target selection.
• The working prototype consisted of 27 targets in 3D Virtual Space and users were asked to point those targets.
• Developed an SVM model to predict/early predict the target in real-time, given the coordinates of the index finger during the flight.
• Using these prediction models, we aim to better understand how users point in Virtual Reality and facilitate early target prediction.

PROJECTS

• Unpaired Image to Image Translation with Cycle-Consistent Adversarial Networks
  Implemented the CycleGAN for the translation of horse images to zebra images and zebra images to horse images.
  
• Review Classification of Amazon Fine Food using Decision Tree and Random Forest
  Decision Tree Classifier, Random Forest Classifier was learned and the least error model was chosen by testing the model with varying hyperparameters.
  
• Neural Machine Translation with Attention
  Implemented a encoder-decoder sequence to sequence model with Bahdanau’s attention for translation of Spanish sentences to English sentences.
  
• Other Projects
  
  1. Built a feed forward neural network (2 Hidden Layered) from scratch to classifiy the MNIST handwritten digits dataset.
     
  2. Built a Canny Edge Detector from scratch in python to detect wide range of edges in images.
     
  3. Implemented a deep neural network for traffic sign classification with an accuracy of 95 percent.
     

---

SKILLS-

C, C++, Python, SQL, HTML, CSS.