AI/ML Engineer at Meta

Developed Python-based distributed LLM training and inference pipelines using PyTorch, Triton, and Kubernetes, supporting 1,500+ enterprise users and improving response accuracy by 18% across multimodal AI workloads. Optimized large-scale inference architecture using TensorRT-LLM, vLLM, quantization, and dynamic batching, reducing inference latency by 27% and lowering infrastructure cost by 19% in production. Built scalable RAG pipelines with Python, FAISS, PyTorch, and embedding frameworks, increasing contextual response relevance by 21% for long-context conversational systems. Engineered distributed training workflows using PyTorch Distributed, FSDP, Megatron-LM, and DeepSpeed for efficient large-model experimentation, checkpoint sharding, and high-throughput parallel training orchestration. Implemented multimodal AI pipelines using TorchMultimodal, transformer architectures, vector embeddings, and evaluation frameworks. Designed robust ML infrastructure with Kubernetes, Docker, Slur