Nikhil Goel

Nikhil Goel

I am passionate about machine learning and am continuously looking for opporunities in this field.

About me

I graduated in 2019 from Birla Institute of Techonlogy and Science, Pilani (Pilani Campus). I am currently working as an associate engineer in Qualcomm, Hyderabad. I have actively worked on projects related to machine learning for the past 2 years and I am actively looking for opportunities in this field.

Computer Science courses completed:

Machine Learning, Data Mining, Operating System, Pattern Recognition, Computer Architecture, Image processing, Data structures and Algorithms, Object Oriented Programming, Digital Design, Microprocessors, Neural networks and Fuzzy Logic

Skills:

Python, Java, R, Git, C, C++, Linux, Assembly language(ASM), Verilog, HTML, Javascript, MySQL

Tools:

Matlab, Github, Azure, IBM SPSS, Microsoft Office, TensorFlow, Bitbucket, Android Studio, MongoDB, Perforce,

Work Experience/Internships

  • Software Engineer, Qualcomm (July 2019 - Present)
    • Working on fixing bugs and feature requests in NAS layer of 4G and 5G.
    • Classification of logs using logistic regression, RNN and Random forest for automation of classification.
  • Intern, Flipkart (Jan 2019 - June 2019)
    • Project on churn prediction and checkout drivers using different ML algorithms and preprocessing techniques..
    • Changes due to checkout project lead to a change in UI in flipkart app and the score generated through churn prediction is tracked a metric. It will be used to treat customers different if they are likely to churn.
  • Data science intern, The Level (May 2018-December 2018)
    • Different projects based on Machine Learning and Data Mining as mentioned below.
    • Got first hand experience about real life problems which can be solved using Machine Learning.
  • Summer Intern, Bhushan Steels Limited, Ghaziabad (May 2017-July 2017)
    • The project was to draw the Single Line Diagram for H&T department in Bhushan steels.
    • Required visiting all the sub departments and gaining good knowledge of their working.

Projects

  • Customer Churn Prediction ( Flipkart)
    • Classification of customers into three classes based on their transactions in a calendar year.
    • Data contained more than 100 variables and about 40 million different transactions hence a lot of preprocessing.
    • Used XGBoost, random forest and LR in boosting ensemble to achieve an accuracy for 78% on 3 class predictors.
  • Checkout Conversion Drivers ( Flipkart)
    • Predict drivers leading to a successful checkout by a customer and their impact on conversion.
    • Used Adobe analytics and Sql queries to aggregate the data of flipkart customers.
    • Used Gradient boosting and Random forest to find feature importance and predict customer conversion.
  • Post Classifier (The Level)
    • Built a topic classifier with the purpose of tagging posts written by users on their news feed.
    • Used Multi-class SVM to classify the topic into relevant topics.
    • Built on reddit posts of the topic as true points coupled with proper noun tagging using spacy library.
    • Achieved an accuracy of 96.3% on dev set making the model ready for live testing.
  • Rate Models (The Level)
    • Built rate models to predict car insurance rates of insurance companies like GEICO.
    • Initial weights of all variables (like car price and age) were gathered from rate filings of insurance companies.
    • Built a similarity score generator using clustering to validate the rate models.
  • Car Price Predictor (The Level)
    • Predicted car price using year, make, model and mileage of the car.
    • Used Gradient Boosting Decision Trees to train the model using the kaggle 1.2 million car price dataset.
  • Financial Signal Processing
    • Time series analysis using Kalman Filters, SVM and Recursive neural networks.
    • Achieved an accuracy of over 95% using open source NYSE data.
  • Movie review text classification (Word sentiment analysis)
    • The movie reviews(Large movie review dataset IMDB) are classified as good or bad using word level features.
    • Used SVM and Naive Bayes to classify the reviews.
  • Image classification
    • Classification on the CIFAR-10 dataset using Deep Convolutional Networks.
    • Used the TensorFlow library in Python to build the network and fine tuned the final layers.