Harsha Vardhan Senior Data Scientist

Address Dubai, UAE Experience 6 Years

Phone +971544611347 LinkedIn linkedin.com/in/harsha-vardhan-data-scientist

E-mail [email protected] GitHub github.com/harsha3187

About
Specialized Data Scientist with a demonstrated experience of 6 years into Machine Learning, Statistical Modelling, Data Mining ,Deep Learning , NLP , Python, Spark ,SQL &
Tableau . Skilled in Data Science process like hypothesis generation, exploratory data analysis, data cleaning, model building, delivering data science pipelines on cloud platforms with
CI/CD pipelines , measure impact of applied models using A/B Testing , results interpretation, and implementation. Good working knowledge on forecasting models, Clustering
analysis/segmentation and recommendation engines.

Skills
Programming Language : Python, Microsoft SQL Server , Hadoop,Java, Spark, Hive, Big Data, Mongo DB, GIT , React and Jira .
Python Packages/IDE’s : Scikit-learn, NLTK, spaCy, Keras, Jupyter, Pandas, NumPy, PyTorch, Matplotlib, OpenCV, imutils, Seaborn, streamlit, pyhive,
SHAP & LIME ,Keras, TensorFlow, TensorFlow Object Detection, ImageAI, Cognitive Services
Cloud Applications : Azure App Services, Azure ML, ML Services, Synapse ,Micro services, Containers, Functions, Azure Data Bricks, Cloud Storage
(BLOBS, Data Lakes, etc.) and Cloud Databases, Azure Cognitive Services, Azure PaaS, Web Jobs, Azure API Management
Service, App Services, and virtualization.

Classification and Regression : KNN, Naive Bayes, Linear Regression, Logistic Regression, SVM, Decision Tree,PCA, Random Forest , XGBOOST, LighGBM,
CATBOOST and Ensemble Models
Time Series and Forecasting : Moving Averages, ARIMA, Facebook Prophet, Exponential Smoothing , Holt’s Winter seasonal Approach.
Text Analytics(NLP) : BOW, TF-IDF, Word2Vec , Doc2Vec, Transformers,Stemming, Lemmatization, Data Cleaning

Clustering : KMeans, Hierarchical ,DBSCAN,K Prototype, Hybrid Models

Deep Learning : Neural Network (ANN ,CNN & RNN), Faster RCNN’s and LSTM, Image and Video Analytics, Background Subtraction

Work History
 
Senior Data Scientist Fragma Data Systems, Dubai, UAE [June 2019 to Present]

Customer Potential (Using Clustering) and Product Recommendations(NPTB):

 Designed and implemented the machine learning models to micro segment the customers to identify the revenue potential , based on
other similar customers in segment (reference) and generate the product recommendations (deals) to relationship managers.
 Developed a decision tree model to micro segment the customers based on industry/needs/purchasing capability/negotiation power/
and other prominent features with the help of decision tree thresholding of independent variables( used as rules).
 Developed ensembled XGBOOST+LIGHTBGM models to generate product recommendations(deals) using financial/non-financial
factors.
 Targeted the revenue of AED 6MN for FY 2021-2022 with a deal acceptance rate of 20% (minimum).
 Deployed the solution using Azure ML, ML Services with Data Lake and Synapse. Used Containers for the deployment.
Fuzzy Matching Utility (Using NLP):
 Designed the fuzzy matching utility to fetch the top N customer matches from 6 sources of customer database sources( Core Banking,
CBRB, Moody’s, Retail Banking, Internet Banking and External Data )
 Used the Word Vectorizers with a custom N gram analyser(splitting the words into mini words) and KNN to perform the search on
sources similarity matrix build using search query( Customer Name).
 Achieved a performance of ~4 secs to search 30 MN customer database for top 10 results .
Safety Video Analytics :
 Developed the safety video analytics solution for manufacturing and construction industries to digitalize the safety system by
identifying the workers without PPE , unsafe activities and workers entering unauthorized sites.
 Developed Faster RCNN’s models for unsafe activity detection and used background subtraction with YOLOV3 for human
detection.
  Used Azure Custom Vision(Cognitive Services) for data tagging and training data generation.
 Deployed the solution using Azure ML, ML Services with Azure Data Bricks , BLOB storage and Functions . Used Containers for
the deployment. Used Azure Web jobs for Video Extraction and sampling.

Predictive Maintenance:
 Designed the predictive maintenance solution to estimate the RUL of a machine using the IOT/sensor data of machine equipment’s.
 Created maintenance and breakdown cycles of a machine by crunching the IOT data of 2 years to derive the RUL.
 Developed Ensembled XGBOOST + Linear Regressor models to predict the RUL of a machine.
 Deployed the solution using Azure ML, ML Services with Azure BLOB storage and Functions . Used Containers for the
deployment. Used Azure Web jobs for data streaming jobs from IOT devices.

Contract Risk Clause Classification:

 Developed NLP based solution to identify the potential risk clauses in contract agreement.
 Used Textract, pyPDF to extract the text from PDF documents and used Azure OCR to extract the text from scanned documents.
 Designed a Multilayer RNN classifier using TFIDF Weighted Word2Vec vectorizer to predict the risk clauses in contract
documents.
 Helped in skimming the documents 3x faster compared to manual skimming.

  Data Scientist Oracle, Bangalore, India [Sep 2017 to Jun 2019]

Branch Sales Forecast and identifying features for implementation:

 Developed sale forecasting model to forecast the sale of operations branch in bank and analysing the possible features/operations/plans
which effects the sale.
 Developed data-gathering and reporting structures from ground up and strategized methods capitalizing on system features.
 Implemented XGBOOST regressor to forecast the sale of a newly formed branch using other relatable branches.
 Performed the continued A/B testing monthly to analyse and improve the features and model attributes.
 Deployed the solution using Azure ML, ML Services with Synapse and Azure DB. Used Containers for the deployment.
Loan Product Recommendation(Next best offer) :
 Developed Product Recommendation system to bank that helps the operations team to suggest the best products to customers and
reduces the turnaround time of response and increase the hit rate(deal conversion rate).
 Implemented the content based collaborative filtering method with KNN(5 Neighbours) to be recommended the products.
 Scaled analytical capabilities across business areas, evolving analytics to influence bank's strategic planning and decision
making.
 Targeted the improvement of deal conversions by 15% (minimum) initially to evaluate the hit rate of the model , by the end of
year model proved to have a hit rate of 23% .
 Deployed the solution using Azure ML, ML Services with Synapse and Azure DB. Used Containers for the deployment.

Data Scientist MetricStream, Bangalore, India [Aug 2015 to Sep 2017]

Insurance Product Recommendation (Next best offer) :

 Developed a recommendation engine based on past customer requirements and successfully engaged products. This model helped in
reducing the human effort of matching loan requirements to a product by 53%.
 Developed Product Recommendation System to the insurance customers based on viewed/selected products using KNN and word
vectorizers with the cosine similarity.
 Implemented Risk Analytics solution and integrated with Global Internal Audit management system to track the risk indicators.
 Helped the strategy and business teams to understand the GRC compliances and set up the risk KPI's based on the historical data.
 Education
  Oct 2011 - Apr Bachelors: Electronics and Communications Engineering
2015 Vikrama Simhapuri University - Andhra Pradesh, India (Graduated with 8.7 GP)

Certifications
Sep 2020 Microsoft certified Data Science Associate