Dheemanth R Joshi
MS (Thesis) Student, Texas A&M University
Agentic AI Systems | Inference Serving & Scheduling | Computer Architecture
Office of the Vice Chancellor and Dean of Engineering, ZACH 5th Floor
462, Peterson Building, Department of Computer Science and Engineering
About
Hi! I'm a Computer Engineering grad student at Texas A&M. I work on the systems that sit underneath AI models: how work gets scheduled, where data is kept, and which machine handles what. Getting those decisions right is usually the difference between an AI system that feels fast and one that doesn't.
These days most of my work is on AI agents, which behave very differently from the workloads our systems were designed for. A single agent task can run for a long time and stretch across many steps, so the usual assumptions about how to schedule and serve them stop holding.
Before Texas A&M I was a pre-doctoral researcher at the Indian Institute of Science, and I studied Electronics and Communication Engineering at PES University, which is where the hardware interest started. Outside of research you'll usually find me on the badminton or pickleball court, so let's rally!
Research Interests
- Agentic AI Systems: serving runtimes for long-horizon, multi-turn LLM agents
- Request & Model Routing: contextual-bandit and cost-aware policies over heterogeneous model pools
- Inference Scheduling: admission control, preemption, and multi-model GPU time-sharing
- KV Cache Management: offload vs. recompute, cache hierarchies, and tiered memory for LLM state
- Computer Architecture: cache replacement, memory systems, and CPU-GPU co-execution
- Operating Systems & Runtimes: resource isolation, process/memory management for AI workloads
News
- Aug 2026 SkipPar is now published in IEEE Xplore as part of the IPDPSW 2026 proceedings.
- Aug 2026 Joined Texas A&M Engineering Experiment Station (TEES) as a Graduate Research Assistant.
- Jul 2026 TRACE-Router: Task-Consistent and Adaptive Online Routing for Agentic AI is now on arXiv: task-trace-granular contextual-bandit routing across heterogeneous model pools.
- Mar 2026 SkipPar accepted at IPDPSW 2026, a hybrid CPU-GPU framework for accelerating LLM training via efficient scheduling of parameter updates.
- Feb 2026 Invited to join as a member of Texas Delta, Tau Beta Pi, The Engineering Honor Society.
- Jan 2026 Joined TEES as a Student Software Developer, building AI-driven education platforms at institutional scale.
- Aug 2025 Started MS (Thesis) in Computer Engineering at Texas A&M University with Graduate Merit Scholarship.
Selected Publications
arXiv preprint arXiv:2607.22465, 2026. Preprint Co-Author
IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW) 2026, 2026. Published in IEEE Xplore, Aug 2026 First Author