At Leidos, I develop BERT-based classification models to organize secure documentation and improve information retrieval for government mission analysts and staff. I also build computer vision models with PyTorch and OpenCV to identify objects in satellite imagery.
I’ve implemented retrieval-augmented generation systems with LangChain and Pinecone, and designed inference services on AWS SageMaker for real-time tactical sensor data. My work also includes automating machine learning pipelines and preparing mission data for model training.
At Together AI, I developed distributed training scripts for large language models on NVIDIA H100 GPU clusters and improved Llama inference latency with FlashAttention and vLLM. I also optimized CUDA kernels and implemented AWQ and GPTQ quantization for more efficient model deployment.

