Bhavya Giri

MLOps/AI Engineer at relevaince.ai. Exploring GPU programming, kernel optimization, and writing C++ on the side.

now

• building ML infrastructure — training pipelines, model serving, deployment on k8s
• learning CUDA — writing custom kernels for transformer inference
• reading about computer architecture and parallel computing

projects

2026	cuda-kernels	custom CUDA kernels for transformer inference — fused attention, quantized GEMM
2025	mlops-pipeline	end-to-end ML pipeline with automated training, evaluation, and deployment
2025	gpu-bench	benchmarking suite for GPU memory bandwidth and compute throughput
2024	vector-db	lightweight vector database with HNSW indexing and CUDA-accelerated search
2024	model-serving	high-throughput model serving with dynamic batching and TensorRT

connect

bhavyagiri.com · twitter · linkedin · email

---

_{// this readme runs on 1 thread. not optimal.}