Manoj Ravindran

Customer Churn Attrition in Banking: Role - Data Scientist.

• Engineered 30+ features from transactional, demographic, and behavioural data to enhance predictive power for customer churn modelling.
• Developed and optimized machine learning models (Logistic Regression, Random Forest, XGBoost) to achieve high predictive accuracy and business interpretability.
• Conducted extensive EDA, feature selection, and hyperparameter tuning to improve model robustness.
• Provided actionable insights to stakeholders, identifying key churn indicators.
• Technologies - Python, Pyspark, Machine Learning, Sparkflows.

Price Gap Analysis on GeM Portal: Role - Data Scientist.

• Designed and implemented a price gap analysis module to detect and validate price variations among similar products on the GeM portal.
• Utilized statistical analysis and Mahalanobis Distance-based outlier detection to identify abrupt price deviations and ensure fair pricing.
• Provided insights on minimum and maximum price thresholds, improving price transparency and preventing anomalies in procurement.
• Technologies - Python, Machine Learning.

LLM-Powered RAG System for Personal Document Search: Role - Data Scientist.

• Designed and implemented a fully local Retrieval-Augmented Generation (RAG) system for private document search using OpenSearch, SentenceTransformers, Ollama LLMs, and Streamlit.
• Built an end-to-end NLP pipeline including OCR extraction (PyTesseract), text chunking, vector embedding generation, and KNN indexing to enable semantic document retrieval.
• Configured OpenSearch hybrid search pipeline combining BM25 keyword matching with vector similarity scoring using normalization and weighted re-ranking for improved search relevance.
• Integrated local LLM inference via Ollama to generate natural-language answers from retrieved document context.
• Developed an interactive Streamlit web application enabling document upload, indexing, hybrid searching, and conversational question-answering over personal PDFs.

Time Series Forecasting on GeM Portal: Role - Data Scientist.

• Data extraction from a MySQL database, preprocessing, encompassing data consistency checks, handling missing values, making time series stationary, and outlier treatment.
• Utilization of a diverse array of forecasting models, including auto ARIMA, SARIMA, smoothing techniques, and generalized additive models (GAM).
• Assessment of model performance through the SMAPE metric, with a configurable system capable of calculating a range of additional metrics such as MAPE, APE, RMSE, R2, MAD, and more.
• The generation of forecasts for the upcoming 12 months' demand using the most suitable model selected through this automated process.