Sourabh Agarwal

Personalized Recommendations for different Personas

Client: Education Sector

Role: Machine Learning Engineer

Project Description: Client wants to build a recommendation enginer that can recommend different content to different personas.

Roles and Responsibilities:

• Used Azure Personalizer Service that is recomending the best content that was getting predicted from Data Science team recommendation model.
• Builded Model Retraining scripts and logging features so that whenever the pipeline runs we can analyze performance.
• Used Azure DevOps and AML to build model retraining pipelines for Recommendation Model(Data Science team has build this model).
• Make CI/CD workflows for automation and testing

Conversational AI

Client: EdTech

Role: Machine Learning Engineer

Project Description: Client wants an smart AI chatbot system that can handle the query of student’s without interfering of teachers/mentors.

Roles and Responsibilities:

• Participated in requirement gathering and technical analysis during the project lifecycle.
• Historic data was provided from client end to perform modelling.
• Have to build 3 different hierarchical based models for client app.
• Used Hugging Face Zero-Shot models to train one of the models and then trained Bert Based Binary classification model.
• Used GPU based cluster from AML studio to train the models.
• All the models and dataset are registered in AML studio.
• Model Scoring File is build and deployed on AKS, as client was expecting 1000 parallel hits.
• Handed over service endpoints to the client for data serving and model predictions.

News Feed Classifier

Client : Investment banker

Project Description: The goal of this project was to

Scrap data from various data source and categorize them according to their type and extracting meta information from articles

Roles and Responsibilities:

• Participated in requirement gathering and technical analysis during the project lifecycle.
• Scraped articles from data sources and make a general logic of scraping by analyses on over 1000 websites
• Builded Scalable Azure Functions that will trigger daily basis and scrap the content using Selenium and Beautiful-Soup.
• More then 60 KPI's were included and calculating in this use-case
• Labeled the collected data for different different categories
• Extracted meta information from articles like company name, date, place etc.
• Use AML studio for development and versioning of ML model and Dataset used for training
• Developed the batch scoring pipeline to take the predictions on daily basis and send it to cosmos
• Entire Lifecycle of Model Building and Deployment was maintained through CI/CD workflows.
• Model Retraining Pipeline was built on AML studio that will be triggered whenever data gets change.

CUSTOMER CHURN PREDICTION

Client: Retailer
Project Description: The goal of this project was to predict and identify the customers who are at high risk of churning out based on historic data

Roles and Responsibilities:

• Participated in requirement gathering and technical analysis during the project lifecycle.
• Analyzed historical sales data using Exploratory Data Analysis and finding out patterns and insights.
• Engineered different features based on the client best knowledge and retail industry.
• Predicted the set of customers who are at a high risk of churning out, with predictive analytics machine learning models based on historic sales data.
• Batch Inference Pipeline is build in Azure Machine Learning studio that will run on every quarter data and store the result under CosmosDB.