Data Science postgraduate, L&T Agentic AI Intern, IEEE-published ML researcher, building GenAI solutions.

github

Bhavyam Ramani  • Peerlist

Built a churn prediction pipeline with SMOTE, hyperparameter tuning and K-Means segmentation. Deployed SHAP-explainable insights via a Streamlit dashboard with Docker containerization, achieved 87% recall.  Applied Kaplan-Meier survival analysis to estimate churn timelines across 4 high-risk customer cohorts.

Customer Churn Analysis

Built a LangChain + FAISS RAG pipeline over government policy documents, enabling accurate, grounded responses to complex scheme queries. Reduced hallucination risk by anchoring all responses to retrieved source chunks critical for policy-sensitive use cases.

Government Policy Intelligence System

Fine-tuned the Ganga-1B LLM on Hindi literature using LoRA (PEFT) for domain-adapted narrative generation, drastically reducing compute cost vs. full fine-tuning. Built a complete Hugging Face Transformers inference pipeline for structured prompt-to-story generation with domain-consistent output.

Hindi Narrative Text Generator

Developed an ML pipeline to predict 30-day hospital readmissions (82% AUC-ROC). Exposed predictions via a Flask REST API with Docker containerization and CI/CD through GitHub Actions.