Raman Choudhary

Working with amazon ads team in building heauristic which will block ip and channel at prebid.There is twitch platform which is basically video gaming platform.We build heuristic which can block ip and channel based on 24 hour data.Apart from that i am also part of buidling pipeline which will aggregate data and create field from raw data and store it in s3.Then it can be used for populating it on quicksight dashboard.We are using various rules based and machine learning technique to build heuristic.

Sr Data Scientist
koch industries, Bengaluru, Karnataka
Working with pricing analytic of molex(sub division of
koch).Building price optimizations model which include
segementing the material based on its attribute and
coming up with new price .There are few project on
forecasting the price based on time series
model,building recommendation engine for price in
various region.

Sr Data Scientist
huawei technologies india pvt ltd, bengaluru, karnatka
AdSearch:
working with adsearch team for creating traning data
so that it can support 16 different langauge.I am
working for vitanameses preprocessing like creating
token,lemma,pos,ner etc for it.For tokenisation we are
trying various technique like standard core nlp,bert
tokeniser etc.For pos we are using bert pos tagging
and some rule based method like deeppavlov which
create rule file and dict file while training and then we
can add some other rule based on our resarch to
handle number,abbreviation,url etc.Since lemma is
same as token in vitanamese ,so no preprocessing on
lemma.

Project 1:

Telco project:

This project aimed at capturing insight from raw location data of user from sprint telecom.We were getting location data of sprint user in every 15 minutes.We were creating many feature from raw location data such as number of visit for each particular stores like walmart,walgreen,target etc.There were other feature like share shift etc.These feature were used for various use cases like stock market prediction of stores.With public dataset already available ,we were adding feature created by us on top of it and verifying whether we can forecast better than public dataset variable only.We had used lstm model for prediction.

Project 1

Time series forecasting:

I was working with aero tool team to build a time series model for labour

cost forecasting.We had biweekly data for various team and we need to clean the data and build model which can predict it for next six month.We started with arima and then we tried fb prophet.And with prophet we were able to reduce the Mape to 8 percent.

Project 2

Aircraft tyre life prediction:

This project aimed at predicting the life of aircraft tyre so that we can know the inventory.There was various sensor which was attached to an aircraft tyres.Whenever the flight lands ,the data get stored in azure sql database.Then we had scrapped the data and created a resource in azure.We did the cleaning and tried various model like linear regression with lasso and random forest regressor .With randomforest regressor we were getting better accuracy.

Project 3

Sentiment analysis:

This project aimed at creating sentiment analysis model on the comment received from leadership people and partner for one of our product.We wanted to see how it is preforming and what additional improvement we can do on top of it by analyzing negative comments.

We did all the text preprocessing and then we tried with various feature engineering technique to convert raw text to feature like bag of word,tfidf,word2vec etc .Then we build logistic regression and svm model on top of it .With logistic regression it was working fine.