ANKIT SHARMA

4+ Years of Experience DataScience

PUNE, India

Summary

I have Around 4 years of extensive experience combining SAS, Python, Front end Automation, Data Processing, Data Visualization, Natural Language Processing, Machine learning, Deep Learning and Reinforcement learning.
I Possess keen interest towards Development using Python.

Languages:

English, Hindi

Favorite Python Packages:

pandas, numpy, scikit learn, keras, statmodels, nltk, gensim, spacy

Experience

Projects:

1) A reinforcement learning Based catalogue Recommender system Duration 8 months [July 2018 – Present])

Catalogue-based recommendations for new tenants facilitate a smooth adoption of the product. To increase customer satisfaction, recommendations based off the new tenant’s Catalogue are surfaced to users. This could potentially minimize the possibility of ad hoc purchases and the time to purchase.

  • Explored and Implement Contextual Bandits algorithms like Epsilon Greedy, Epoch Greedy, LinUCB, and Thomson sampling algorithm for new and existing users to the system.
  • Implement Reinforcement Learning based model Markov chains, Markov Decision Process, Q learning Model free for improving quality of recommendations.
  • Validation of quality of recommendations using several measures likes RMSE, MAP, NDCG, Cumulative regret, CTR etc.
  • Handling calls with the customer and deploying code in Object oriented format using PEP8 and PEP257 standard.
  • 2) Deep Neural Networks compression: Xoriant Solutions [March 2018-June 2018]

    Neural networks are both computationally intensive and memory intensive, making them difficult to deploy on embedded systems with limited hardware resources. To address this limitation, we implemented “deep compression”, a three-stage pipeline: pruning, trained quantization and Huffman coding, that work together to reduce the storage requirement of neural networks by 35× to 49× without affecting much of their accuracy.

  • Implement Quantization and Huffman coding using TensorFlow.
  • Model was tested on VGG-16, Resnet and Google net.
  • 3) Sentiments analysis of form 10K for Portfolio Analysis (Duration 6 months [Mar 2017 – October 2017])
  • A Form 10-K is an annual report required by the U.S. Securities and Exchange Commission (SEC) that gives a comprehensive summary of a company’s financial performance. The idea is to see if the 10k forms filled by companies have any predictive power in forecasting the movement of stock markets.
    • Scrapped form 10K from EDGAR website using Beautiful Soup library and Extracted required Meta Data like Filled Date, Company Name, Section 7A, Power of attorney from it.
    • Getting Adjusting closing prices of stocks using Quandl API.
    • Data Preprocessing, Manipulation and Feature Extraction using Re, Pandas, Numpy, NLTK, Gensim modules. Predicted Sentiments using Recurrent Neural
    Networks (RNN) with LSTM.
    • Implemented and use of NLP concepts like POS tagging, Name Entity Recognition, Word2vec modelling, Word Embedding, Topic Modeling, Entity Relationship extraction etc.
    4) Financial Risk Prediction for Insurance Plans (Duration 3 months [ Jul 2016 – October 2016])
    Implemented a predictive model to predict Financial Risk on scale of 10 involved in launching a Insurance Plan. Analyst can feed data to model and can train accordingly.
    • Data Preprocessing, Manipulation and Feature Extraction is done using Pandas, NumPy, Re modules in Python.
    • Implemented Predictive modeling using various classification Algorithms like Logistics Regression, Decision Trees, Random Forests and SVM
    5) Information Extraction: Infosys Technologies [ June 2015-June 2016]
    Implemented a project to Extract Desired Information from various PDF files, Databases and manipulating it to prepare a final report as required by ETL team.
    • Involved in writing Regular Expressions for extracting information like SSN, Contributions, Date of Birth etc from raw data.
    • Involved in Slicing and Dicing of Data as per the business logic.
    • Extracted Information from various tables and produced a final report as required by ETL team.
    • Developed scripts in PYTHON for automation of reports and manual work flows using sikuli libraries. Sikuli automates anything seen on the screen. It uses image recognition to identify and control GUI components.
    • Automation of multiple workflows results in saving average of 100 hours monthly.
  • Skills

    Artificial Intelligence, Machine Learning, Natural Language Processing

    Joined: February 2019