Ragapriya Kovvuri

Architected a Generative AI shopping assistant using Amazon Bedrock, LLMs, Retrieval-Augmented Generation (RAG), and vector search to enable conversational product discovery and improve recommendation relevance by 32%. Developed hybrid retrieval pipelines combining semantic search, BM25, product metadata filtering, embeddings, and cross-encoder reranking, increasing search precision by 35% and reducing irrelevant recommendations. Built scalable ingestion and indexing pipelines using Python, Spark, SageMaker, and OpenSearch to process millions of product listings and cut indexing latency by 45%. Engineered intelligent recommendation pipelines leveraging browsing history, purchase behavior, contextual signals, and LLM-based reasoning, increasing click-through rates by 18% and conversion rates by 12%. Implemented automated LLM evaluation frameworks (RAGAS, LangSmith) and offline datasets to measure answer quality, groundedness, retrieval effectiveness, and hallucination rates, reducing un

Architected AI solutions using PyTorch, TensorFlow, and Transformer-based models to automate digital asset tagging, metadata generation, and semantic content understanding, improving asset discoverability by 32%. Developed computer vision and NLP pipelines for image classification, object detection, OCR, and keyword extraction, reducing manual tagging effort by 60% and improving metadata accuracy by 30%. Built semantic search systems leveraging vector embeddings, FAISS, Elasticsearch, and hybrid retrieval, reducing asset search time by 40% and increasing search relevance by 28%. Implemented end-to-end MLOps pipelines using MLflow, Docker, Kubernetes, and Azure ML, automating model training, deployment, monitoring, and retraining while reducing release cycles by 50%. Integrated AI-powered tagging and semantic search capabilities into AEM and DAM solutions within Adobe Experience Manager, improving enterprise content retrieval efficiency by 35%.