Parijat Das

Dynamic ML engineer with experience in training various neural networks including multimodal models, gnn, llms and building ML systems like RAG, search engine, recommender systems etc.

· Designed a RAG application with LlamaIndex framework, Pinecone Vector DB and Open-source embedding models from Huggingface to query documentations from S3 bucket and made a Streamlit based chat application, resulting almost 15% production boost in team level.

· Developed an advanced computer vision tool integrating layout detection, optical character recognition (OCR), and natural language processing (NLP) techniques to automatically identify unethical promotions and messages in product images, significantly reducing the need for human intervention and increased true positive flagging rate by 10%.

Developed a fine-tuned multimodal embedding model with supervised contrastive representation learning on an Amazon product data to detect fake product descriptions by calculating similarity scores between product titles/descriptions and corresponding images.