Harsh Mehta
Machine Learning Engineer at Casetext- Claim this Profile
Click to upgrade to our gold package
for the full feature experience.
Topline Score
Bio
Credentials
-
Machine Learning
CourseraSep, 2018- Nov, 2024
Experience
-
Casetext
-
United States
-
Technology, Information and Internet
-
1 - 100 Employee
-
Machine Learning Engineer
-
Jul 2021 - Mar 2022
● Conceptualized and developed an online testing evaluation system to production to determine which search models work better as determined by users ● Developed a search ranking model that outperformed the production models based on test sets by 3% ● Presented latest NLP research to machine learning team on topics such as BERT, COIL, top2vec ● Created new models to return ranked lists of the most relevant long spans of text (paragraphs) for a search query ● Created a pipeline to… Show more ● Conceptualized and developed an online testing evaluation system to production to determine which search models work better as determined by users ● Developed a search ranking model that outperformed the production models based on test sets by 3% ● Presented latest NLP research to machine learning team on topics such as BERT, COIL, top2vec ● Created new models to return ranked lists of the most relevant long spans of text (paragraphs) for a search query ● Created a pipeline to cluster documents into relevant topics in an unsupervised manner Show less ● Conceptualized and developed an online testing evaluation system to production to determine which search models work better as determined by users ● Developed a search ranking model that outperformed the production models based on test sets by 3% ● Presented latest NLP research to machine learning team on topics such as BERT, COIL, top2vec ● Created new models to return ranked lists of the most relevant long spans of text (paragraphs) for a search query ● Created a pipeline to… Show more ● Conceptualized and developed an online testing evaluation system to production to determine which search models work better as determined by users ● Developed a search ranking model that outperformed the production models based on test sets by 3% ● Presented latest NLP research to machine learning team on topics such as BERT, COIL, top2vec ● Created new models to return ranked lists of the most relevant long spans of text (paragraphs) for a search query ● Created a pipeline to cluster documents into relevant topics in an unsupervised manner Show less
-
-
-
Lowe's Companies, Inc.
-
United States
-
Retail
-
700 & Above Employee
-
Data Scientist
-
Jun 2020 - Jun 2021
● Created short-term daily sales forecast for each department across Lowe’s stores, using GBDT and GLM forecasting methods, that consistently outperformed the accuracy of the sale’s plan ● Tested and optimized model hyperparameters to improve model accuracy ● Helped integrate running operations on a distributed platform using spark (PySpark and koalas) ● Tested and developed methods to forecast on targets with very little historical data ● Contributed to existing Lowe’s… Show more ● Created short-term daily sales forecast for each department across Lowe’s stores, using GBDT and GLM forecasting methods, that consistently outperformed the accuracy of the sale’s plan ● Tested and optimized model hyperparameters to improve model accuracy ● Helped integrate running operations on a distributed platform using spark (PySpark and koalas) ● Tested and developed methods to forecast on targets with very little historical data ● Contributed to existing Lowe’s forecasting tool by adding new objective function in Huber loss Show less ● Created short-term daily sales forecast for each department across Lowe’s stores, using GBDT and GLM forecasting methods, that consistently outperformed the accuracy of the sale’s plan ● Tested and optimized model hyperparameters to improve model accuracy ● Helped integrate running operations on a distributed platform using spark (PySpark and koalas) ● Tested and developed methods to forecast on targets with very little historical data ● Contributed to existing Lowe’s… Show more ● Created short-term daily sales forecast for each department across Lowe’s stores, using GBDT and GLM forecasting methods, that consistently outperformed the accuracy of the sale’s plan ● Tested and optimized model hyperparameters to improve model accuracy ● Helped integrate running operations on a distributed platform using spark (PySpark and koalas) ● Tested and developed methods to forecast on targets with very little historical data ● Contributed to existing Lowe’s forecasting tool by adding new objective function in Huber loss Show less
-
-
-
University of South Carolina
-
United States
-
Higher Education
-
700 & Above Employee
-
Graduate Teaching Assistant
-
Aug 2014 - Jun 2020
● Teach up to 38 students, introductory math courses such as precalculus, finite mathematics and calculus, covering concepts like limits, differentiation, linear regression and matrix multiplication ● Organize group worksheets, helping students communicate and accurately frame their ideas ● Make and grade exams and quizzes to evaluate student performances ● Teach up to 38 students, introductory math courses such as precalculus, finite mathematics and calculus, covering concepts like limits, differentiation, linear regression and matrix multiplication ● Organize group worksheets, helping students communicate and accurately frame their ideas ● Make and grade exams and quizzes to evaluate student performances
-
-
-
Algomatrix
-
Mumbai
-
Analyst
-
Jun 2019 - Aug 2019
● Gathered data about various equities and derivatives, cleaned the data and made a user friendly program to call directional trades, straddle trades and strangle trades ● Faced with the problem of classification where representing the smaller classes with precision was more important than having high accuracy ● Resolved the above by implementing an original version of boosting to create linear combinations of features that capture the under-represented classes better ● Used a random… Show more ● Gathered data about various equities and derivatives, cleaned the data and made a user friendly program to call directional trades, straddle trades and strangle trades ● Faced with the problem of classification where representing the smaller classes with precision was more important than having high accuracy ● Resolved the above by implementing an original version of boosting to create linear combinations of features that capture the under-represented classes better ● Used a random forest with these new features Show less ● Gathered data about various equities and derivatives, cleaned the data and made a user friendly program to call directional trades, straddle trades and strangle trades ● Faced with the problem of classification where representing the smaller classes with precision was more important than having high accuracy ● Resolved the above by implementing an original version of boosting to create linear combinations of features that capture the under-represented classes better ● Used a random… Show more ● Gathered data about various equities and derivatives, cleaned the data and made a user friendly program to call directional trades, straddle trades and strangle trades ● Faced with the problem of classification where representing the smaller classes with precision was more important than having high accuracy ● Resolved the above by implementing an original version of boosting to create linear combinations of features that capture the under-represented classes better ● Used a random forest with these new features Show less
-
-
-
University of Michigan
-
United States
-
Higher Education
-
700 & Above Employee
-
Grader
-
Jan 2013 - Dec 2013
● Grader for ‘Introduction to Number Theory’ (Math 575) under Dr. Lagarias, assisting 20 students per week ● Grader for ‘Fourier Analysis’ (Math 375) under Dr. Montgomery, assisting 20 students per week ● Helped improved course quality through regular feedback to instructors about needs of students to be addressed ● Grader for ‘Introduction to Number Theory’ (Math 575) under Dr. Lagarias, assisting 20 students per week ● Grader for ‘Fourier Analysis’ (Math 375) under Dr. Montgomery, assisting 20 students per week ● Helped improved course quality through regular feedback to instructors about needs of students to be addressed
-
-
-
Ventura
-
India
-
Capital Markets
-
700 & Above Employee
-
Volunteer
-
Jun 2010 - Aug 2010
Used stochastic models from Spider8 software and the Elliott wave principle to project behavior of commodities Used stochastic models from Spider8 software and the Elliott wave principle to project behavior of commodities
-
-
Education
-
2014 - 2020
University of South Carolina
Doctor of Philosophy - PhD, Mathematics -
2010 - 2013
University of Michigan
Bachelor of Science - BS, Mathematics
Community
