Niraj Kumar

Research and Development in GenAI Applications. Worked on 100M synthetic dataset generation ensuring data quality and consistency for machine translation. Proficient at building robust Machine Learning, Recurrent Neural Networks (RNN) and Deep learning models using Transformers. Design and develop state-of-the-art deep learning models for machine translation in codemix language (i.e. Hinglish). Identify and assess available different parallelisation techniques (DP, DDP, FP16) for GPU optimizations. Performed model validation and benchmarking using BLEU Scores, Semantic Similarity, and Human Annotations. Introduced a new novel method for data generation, which resulted into 30% better quality, and at par with GPT-4 Datasets. Build Interactive UI for internal Translation model performance testing using Flask, JS, HTML, CSS, Figma. Build Query and Response system for client POCs which can cater to Hindi, English, and Hinglish languages using our in-house SOTA models and OLLAMA/Langchain interface. Other application works in NLP - NER, POS tagging, and Sentiment Analysis.

Managing complete marketing graphics work for an Email Marketing Company named Mailmodo. Build Email Templates, Landing Pages and Marketing Banners and videos. Learn collaborative work between marketing, product, and partnership team.