Vankdoth Vyshnavi

Streamlit Developed an AI-powered You Tube Research Assistant using Python and Streamlit that extracts You Tube transcripts and converts them into a searchable knowledge base. Implemented text chunking and generated semantic embeddings using Hugging Face Sentence Transformers for representing transcript content as numerical vectors. Built a FAISS vector database with Lang Chain and implemented similarity-based retrieval to identify relevant transcript sections for user queries. Integrated Llama 3.2 through Ollama to generate context-aware responses grounded strictly in the retrieved transcript content. Implemented multi-turn conversation history and contextual follow-up question handling, allowing users to continue discussions about the analyzed video. Added an "I don't know" fallback mechanism when sufficient information is unavailable in the transcript and developed a professional Streamlit conversational interface. Credit Card Fraud Detection Python, Pandas, Scikit-learn, Random Forest, SMOTE Built a classification system to detect fraudulent credit card transactions from a highly imbalanced transaction dataset. Handled class imbalance using SMOTE and under-sampling, then trained Logistic Regression and Random Forest models with Randomized Search CV tuning. Evaluated performance using confusion matrix, precision, recall, ROC-AUC, and Precision-Recall curves with focus on fraud detection. Movie Recommendation System Python, NLP, Scikit-learn, Streamlit, TMDB API Developed a content-based movie recommendation system using TMDB 5000 Movies and Credits datasets by combining genres, keywords, cast, crew, and overview into a unified feature. Applied text preprocessing, stemming, and Count Vectorizer with 5,000 features to represent movies as numerical vectors and calculated cosine similarity. Implemented a recommendation function to return the top 5 similar movies and integrated the TMDB API with Streamlit to dynamically display movie posters.