Akhil Shekkari
AI engineer working on LLM inference, GPU systems, and agents.
I learn systems by rebuilding them, profiling them, and writing down what actually happened. My current work focuses on KV-cache management, continuous batching, tool-using agents, and evaluation.
Selected work
Mini-vLLMPaged KV cache, block allocation, and continuous batching
Coding AgentPlan-then-execute CLI with repository-scoped tools
Agent FrameworkAn explicit tool loop with MCP integrations
All work →
Recent writing
Lessons I Learnt from Building vLLM Internals
Context in LLMs: what determines it and what it costs
GPU Fundamentals & LLM Inference Mental Models
All writing →
Background
MS in Applied Machine Learning at the University of Maryland. Previously worked on a Pfizer R&D project at Atrium and spent three years building ML and data systems at Tezo. I care about correctness, honest benchmarks, and clear limitations.