SURAJ SUVENDU CHATTERJEE

Summary

Developed and deployed end-to-end ML models for credit risk scorecards on time series data for Indian Banks & NBFCs (Application Scorecard, Behaviour Scorecard, Collection Scorecard, etc) with a strong focus on predictive modeling, explainability, and automated deployment on AWS. Achieved a ~90% AUC metric for Behavior Scorecard predicting the probability of default in the next 3 months. Optimized and implemented multiprocessing in data processing scripts which increased processing speed by 90% Led and managed a cross-functional team of 5 engineers, overseeing end-to-end delivery of multiple ML and AI projects—from initial planning and architecture design to model development, deployment, and post-launch optimization. Automated data & model pipelines and model scoring systems using AWS Lambda, S3, and SageMaker, improving model turnaround and scaling capabilities. Processed and analyzed large-scale financial and behavioral datasets using Pandas, NumPy, and SQL, enabling robust feature engineering and anomaly detection and showcased the insights via visualizations for client and stakeholder's understanding. Gained deep familiarity and implemented statistical ML models using PyData Stack (Pandas, Numpy, Scikit-Learn), SQL, Bayesian Hyperparameter Tuning (Optuna), Ensemble Boosted Trees (LGBM, XGBoost), Feature Selection (RFE, Boruta) and deployment technologies such as RESTful Django, FastAPI, Flask, Docker & AWS and experimented with TensorFlow and PyTorch for advanced use cases. Led various experiments in deep learning, while continuously optimizing performance through hyperparameter tuning. Authored detailed internal documentation, model explainability reports, and stakeholder presentations showcasing model insights, risk stratification, and performance over time. Contributed to the adoption of AI-first approaches within the organization by identifying use cases for automation and Generative Al experimentation.

Summary

Obtained data through Web Scraping using Selenium, Scrapy, and BeautifulSoup from multinational banks' websites Maintained and preprocessed the data to remove anomalies, treat outliers, etc for further analysis using Python Performed Statistical, Contextual (using TFIDF, IDF & Glove), and Sentiment analysis (using AFINN, Vader) through Natural Language Processing (NLP) to see how Multinational Banks are supporting the Sports for development sector Created interesting visualizations on the processed data like Wordclouds, Pie Charts, etc. using Seaborn and Matplotlib libraries Designed a Bitcoin price prediction model using Deep Learning techniques such as LSTM, RNN, etc, and statistical modeling techniques like ARIMA Performed sentiment analysis on Bitcoin news articles using unsupervised Sentiment analysis techniques such as Vader, Textblob, AFINN, etc., and incorporated the same into the Bitcoin deep learning model

Summary

Designed and implemented Generative AI solutions, including LLM-based intelligent chatbots using RAG (Retrieval-Augmented Generation) architecture with AWS Bedrock, Llama2, Claude, GPT and vector databases (FAISS). Developed scalable ML pipelines for traditional ML, NLP and GenAI models using Amazon SageMaker, Bedrock, EC2, Lambda, ECS, ECR, S3, Terraform and Snowflake adhering to MLOps best practices for versioning, CI/CD, and model monitoring. Integrated serverless backend APIs for AI services using AWS Lambda and deployed them using API Gateway and Terraform. Contributed to real-time AI monitoring dashboards and performance reporting using Python-based visualization libraries, Streamlit and Amazon CloudWatch logs. Explored and implemented NLP, AI & LLM techniques in NLP tasks like entity resolution, question answering, text summarization, and semantic search, with a focus on scalability, latency, and response accuracy. Embraced MLOps workflows, using Github Actions, Docker and Terraform for containerization and infrastructure provisioning, and monitored deployments through CloudWatch. Regularly presented technical findings and PoC results to stakeholders and cross-functional teams through clear visualizations and demos.