Senior Data Scientist
CurrentNLP, Generative AI, and Embedding Systems:Developed an embedding-based Retrieval Augmented Generation (RAG) architecture for Verizon Business use case, integrating it with large language models such as LLaMA-2 and Falcon for enhanced question-answering capabilities. Used Neo4j platform to build a knowledge graph to support these models.Fine-tuned large language models like FLAN-UL2, GPT-J for call summarization and topic segmentation use-cases. Used model quantization, low rank adapters (LoRA), batched streaming, and runtime environments (TorchScript, ONNX, TensorRT) for efficient model deployment.Deployed semantic similarity measures, including S-BERT, instruction fine-tuned text embedding models (INSTRUCTOR, FlagEmbedding), and LangChain to optimize semantic search performance.Computer Vision:Used Masked R-CNN, SegFormer, and YOLO for real-time product segmentation. Trained an image similarity system using convolutional neural network and vision-transformer feature extraction and contrastive learning of similar images.
