Souradeep De

Data Scientist with more than 3 years of experience over Predictive Modeling, Data Processing, Natural Language Processing, Generative AI and Deploying highly scalable Machine Learning models. Passionate about NLP & Deep Learning. Proficient in Python, Statistics, Pre-processing, Visualization, SQL Databases, and detail-oriented. Excellent communicator, deadline-driven, and adept at issue resolution.

Project 1 - SmartQGen: AI-Powered Question Generation and Auto-Answering

Roles & Responsibilities: -

• Developed a text similarity matching model suggesting user questions based on their input.
• Created a questionnaire database with scoring against similar sentences.
• Trained 'distil-roberta-base' model for text similarity.
• Achieved Spearman Correlation of 0.87 and Pearson Correlation of 0.88 after complete training.
• Stored questionnaire database in Azure Cosmos-db.
• Developed REST APIs using Azure Functions for question suggestions and model retraining.

Project 2 - CreditGuard: Predictive Credit Scoring System for Banks

Roles & Responsibilities: -

• Developed predictive model for default prediction and calculated credit scores using logit regression, WOE, and IV
• Conducted data analysis and preprocessing, handled missing values, and outliers
• Performed feature transformation, selection, and model optimization
• Developed credit scoring model based on odds and coefficients

Project 3 - Emotionise: AI-Driven Content Generation and Emotion Detection

Roles & Responsibilities: -

• Developed emotion detection, question generation, and tool-tips generation NLP models using LLMs (GPT-3).
• Cleaned and preprocessed data, configured BERT transformer
• Trained and validated models, deployed on Azure VM

Project 4 - Blood Pressure Abnormality Prediction Model

Roles & Responsibilities: -

• Collected essential features from domain experts
• Cleaned and transformed data, handled multicollinearity
• Conducted feature selection and model optimization.
• Upon hyperparameter tuning and using, cross-validations got a test accuracy of 88%.
• Deployed on Azure VM. Developed front-end using Streamlit.

Project 5 - Keyword Extraction and Website Industry Prediction

Roles & Responsibilities: -

• Scraped website landing pages using Python and BeautifulSoup to extract keywords.
• Retrieved historical metrics such as search volume and CPC for the keywords via Google Ads API.
• Filtered and selected top 20 keywords based on metrics for further analysis.
• Trained GPT-3.5-turbo-instruct model on extracted keywords to identify the most likely industry of the website.
• Deployed the model on a cloud platform for real-time industry classification predictions.
• Applied data pre-processing and model optimization techniques to enhance accuracy and efficiency.

• The Data Science Course 2020: Complete Data Science Bootcamp - [28/09/2020].

Certificate Number : UC-6c0b3b3b-c998-4f33-8184-d10ec9fe4725

Certificate URL : http://ude.my/UC-6c0b3b3b-c998-4f33-8184- d10ec9fe4725

• Databases and SQL for Data Science with Python by IBM - [23/03/2022]
• Introduction to Machine Learning in Production - [09/04/2023]

• Led team to achieve an accuracy of 85% from 70% for a text extraction model, earning recognition from upper management and financial reward.
• Improved delivery of Textual Intelligence models by using open-source, pre-trained transformers and Large Language Models, realizing overall increase in customer satisfaction and cost efficiency.
• Consistently maintained high customer satisfaction ratings.

• Member of St. Thomas' College of Engineering and Technology Alumni Association
• Captain of St. Thomas' College of Engineering and Technology Alumni Table-Tennis team (2021-present)