Raman Lok Sainadh Reddy

B.S.(Hons.) in Mathematics and Computing

Projects

Document Intelligence Chatbot using RAG, FAISS, and LLMs

Developing RAG based chatbot to answer PDF-based queries using semantic retrieval and LLM-generated responses

  • Building a document pipeline for text extraction, chunking, embeddings, and vector indexing using FAISS
  • Implementing transformer embeddings and FAISS similarity search for context-aware document retrieval analysis
  • Integrating LLM-powered generation with retrieval grounding to reduce hallucinations and improve relevance

SmartLink: URL Shortening & Analytics Platform

Built a FastAPI-powered URL shortening platform with authenticated link creation, redirection, and analytics

  • Implemented Base62 encoding, collision handling, and indexed lookups to achieve sub-100ms URL resolution
  • Developed click-stream analytics pipelines generating daily and weekly traffic insights across shortened URLs
  • Designed REST APIs and normalized MySQL schemas for scalable URL storage, retrieval, and user management

Context-Aware Next Word Prediction using LSTM

Generated 72K+ text sequences via sliding-window context extraction from unstructured textual language modeling

  • Learned embeddings across 8.4K+ unique tokens for context-aware neural language modeling and prediction
  • Tuned sequence length and stacked LSTM's, improving validation accuracy (5.8% to 8.4%) through experimentation
  • Produced top-3 next-word recommendations through softmax inference, achieving 19.6% top-3 accuracy

Interpretable Multi-Class Loan Default Risk Prediction

SHAP, Anova, VIF Analysis

  • Engineered 80+ high-dimensional features across 51k+ loan records to optimize Multi-class default risk modeling
  • Mitigated multicollinearity across 80+ variables via VIF Analysis, Anova, Chi-square Tests isolating 35 indicators
  • Interpreted key delinquency patterns across 4 risk tiers using SHAP, ensuring transparent algorithmic decisions

Customer Churn Prediction and Retention Analytics using SHAP

SMOTEENN, PCA

  • Engineered churn prediction models, improving Macro F1-Score from 72% to 91% through imbalance handling
  • Addressed severe class imbalance through SMOTEENN based hybrid resampling, noise reduction, and PCA analysis
  • Mitigated categorical class imbalance, improving the Minority class algorithmic precision by 21% (68% to 89%)

Skills

PythonCC++SQLHTMLCSSTime Series ForecastingNumPyPandasMatplotlibScikit-learnXGBoostTensorFlowKerasJupyter NotebookVS CodePower BIExcelAnacondaGitHubCanvaFigmaMySQLPredictive ModellingMachine LearningDeep LearningStatistical AnalysisNatural Language Processing (NLP)Ensemble LearningModel InterpretabilityRNNsLSTMsTransformers

Education

IIT Kharagpur

Sri Chaitanya Junior College

Sri Chaitanya High School

Built with Round Zero