Writing – Tanmmay Kanhaiya

• High-Throughput LLM Inference: Understanding vLLM and PagedAttention

  Dissecting how vLLM revolutionizes large language model serving through PagedAttention, non-contiguous KV-cache management, and continuous batching.

• Building Alpha-Ray: An Intelligent Ray Budget Allocator for Path Tracing

  Designing an ML-guided ray budget allocator to cut real-time ray-tracing cost by predicting per-tile sample counts from G-buffer features.