Asper.ai

Client - COTY
Developed a Long Term Forecasting method(~4 years) by using simple and tree-based models at an overall category level for Consumer Beauty with an accuracy of 95%.

Came up with a SHAP-based factor contribution method to explain the difference in predictions at the YOY level

Client-AVON
The project aims to do short-term forecasting(for up to 10 campaigns in the future) at the Profile-SKU level for Beauty and Fashion products

Developed an approach for classifying the profiles into Existing vs New based on the campaign duration
Used an attribute-based modeling method for predicting the demand for new products

Used an Auto ML framework to generate predictions for Existing and New Products to beat the operational forecast by ~5% at a combined level

Data Scientist

Equifax

Involved in the thorough validation of the entire data cleaning, imputation and creation methods

Did a sample bias analysis to check whether the XGBOOST model is biased towards a certain age group

Recreated the XGBOOST model on BigQueryML while removing the age feature to check the bias and performance in terms of the capture rates.

Accolite Digital

Generated Insights which helped the customer to 
understand the market behavior and its effect on 
the jobs considering the weather factors.

Created Log Based Linear Model at Monthly level 
for different states in US which is able to predict the 
jobs with an approximate accuracy of 80%
Came up with a XGBOOST Model which gives the 
client a simulation of how to divide the marketing 
spend at weekly level.

Designed a slope based optimization approach 
which gives the client an efficient way to divide the 
overall marketing budget across the four marketing 
categories and helps to achieve improved ROAS 
Predicted the cancellation rate for the number of 
leads converted using Random-Forest Model 
along with Hyperparameter Tuning using optuna 
for each of the individual states which helps the 
client to adjust its capacity

Developed a saturation point approach that will
help the client to know the max budget which can
be spent across a particular month and the maximized
number of jobs across each of the marketing channels

This first step of this project is about extracting details from resumes such as email-id, Years of Experience, and the Number of companies Changed.
 
First filtering is provided based upon a rule-based matching which was discussed by the accolite internal team based upon the above factors. 

The next step was to implement a method that will take into consideration the context similarity between the resume and the job description provided For implementing the context similarity method we tried out different approaches but narrowed it down to Siamese Network. 

We took the embeddings of the sentence using Sentence Bert and on top of that built a Siamese network that can predict the similarity between the resume and the job description.

Data Science Intern

Data Scientist with an overall 2.5 years of Experience. My work includes understanding the user requirements and framing them up into a machine learning problem. I have experience in creating classical machine learning models which include Random-Forest, XGBOOST, SVR and Log-Linear Models. I also have experience in creating custom features that add value to machine learning models. Also equipped with using different Feature Selection methods which helps the model to use relevant features

Taplio

vishal sharma's Linkedin Analytics

vishal sharma

Want to drive more opportunities from LinkedIn?

Taplio

.css-1jgreyo{background-image:linear-gradient(284deg, rgba(39,43,72,1) 0%, rgba(33,47,183,1) 100%);color:transparent;-webkit-background-clip:text;background-clip:text;}vishal sharma's Linkedin Analytics

vishal sharma

Want to drive more opportunities from LinkedIn?

vishal sharma's Linkedin Analytics