Cummins Inc has worked on data science initiatives to bring down costs related to repair and maintenance. I have set up the data pipeline which reads data from various tables and creates a single dataset which helps in building machine learning models to determine probability of component failure given a symptom and service model name. This helps in maintaining appropriate stock at various warehouses.Second use case is to find the reasons of repeated visits. Repeated visits are source of revenue impact as warranty is breached when an engine comes for repair within a month of earlier visit. This is a data discovery exercise where I am writing R Script to generate dataset which contains reason, symptoms related to a repeated visit and find probable causes.

Set up and matured Data Science practice for IOTWorKS.Set up cloud team by putting the team together and leading the development of a solution on Bluemix cloud platform.Worked on a number of RFPs related to Data Science proposals.Anomaly detection for unlabeled data (unsupervised learning). This solution has become a high value proposition and is shown as the first important artefact for visiting customers.Manufacturing Analytics - Pilot for Manufacturing client to gather insights from data using exploratory data analysis. This was targeted to identify production lines with good OEE (Overall Equipment Efficiency) and ways to maximize it.Designed a predictive maintenance showcase for Aircraft maintenance. Data Science related use case was defined, designed and executed by me using R Statistical Package. This demo has been presented in International Airshow at Farnborough, UK.Implemented Predictive maintenance for Connected Products using R Programming language for condition monitoring. Combined device and asset data to create a labeled dataset which was later used to determine probability of failure of a component in the next fifteen days.

Application of Transfer Learning in various use cases. Currently working on detecting diseases in an XRay by using
Resnet architecture and fine tuning with provided dataset. Productionizing models on AWS Lambda platform with
Python starlette package. Have successfully applied transfer learning on Geospatial images and finding existence of
various objects.
Technical Environment: Pytorch, Fastai, AWS

Condition monitoring by analysis of Vibration data for determining the health of a device. Time waveform, Spectrum
Analysis, Fourier Transform are some of the techniques used here. Designing rules based on statistical considerations
such as movement of orders, persona changes, etc. Setting up the architecture for scaling of Fault Detection,
interpretation of faults, etc. Further suggested and developed framework for predictive models to determine the
probability of failure of a component.

Prediction of alpha based on sentiment and stock price movement data. Processing of financial articles and
determining sentiments attached to it, using NLTK library (sentient and vader). External sentiment data was
correlated with Stock movement data and subsequently, machine learning models were built. These models were
made accessible using Flask in Python. Also a search mechanism was developed using TF-IDF models and cosine
similarity score. Given a stock and sentiment polarity, most relevant articles related to stocks are found. Models
created were operationalized with help of Flask library available in python.

Worked on data science initiatives to bring down costs related to repair and maintenance. I have set
up the data pipeline which reads data from various tables and creates a single dataset which helps in building
machine learning models to determine probability of component failure given a symptom and service model name.
This helps in maintaining appropriate stock at various warehouses.
Second use case is to find the reasons of repeated visits. Repeated visits are source of revenue impact as warranty is
breached when an engine comes for repair within a month of earlier visit. This is a data discovery exercise where I am
writing R Script to generate dataset which contains reason, symptoms related to a repeated visit and find probable
causes.