Subhash Mothukuru

Designed, trained, and fine-tuned transformer-based LLMs (GPT, BERT, LLaMA) for NLP tasks; built automated ML pipelines with Python, TensorFlow, PyTorch, and Hugging Face; implemented real-time data ingestion with Spark and Airflow; deployed MLOps workflows on Azure, enabling reproducible training, monitoring, and large-scale production deployment; optimized inference with ONNX, TensorRT, FastAPI; implemented vector databases (Pinecone, FAISS) with Retrieval-Augmented Generation (RAG) to enhance semantic search across 5+ enterprise products.

Developed and fine-tuned large language models (LLMs) using PyTorch and LLaMA 2 for generative ad copy; built vision diffusion models (Emu, EmuEdit, EmuVideo) for image/background generation and animated creatives; conducted prompt tuning and RLHF experiments; engineered real-time ad recommendation models (GEM) contributing to lifts in conversions; designed scalable inference pipelines with ONNX, TorchScript, and AWS; led canary deployments and A/B testing; built multi-modal pipelines and integrated AI features via RESTful APIs.