Language Models as Systems Operators
Can an LLM tune a storage engine end to end, and where exactly does that capability stop?
- HotStorage '24 — Best Paper
- SYSTOR '26
- ELMo-Tune · TellyTune
Ph.D. Candidate · Arizona State University
Viraj Thakkar
I teach databases new tricks and poke at storage engines until they spill their secrets.
- Key-Value Stores
- Filesystems & Caching
- Disaggregated Infrastructure
- Storage for AI/ML
Hi, I’m a perpetually curious person who gets way too excited about taking big, intimidating systems apart (sometimes literally) to see what makes it tick. Then, on a good day, I put them back together a little bit better than I found them.
- Best Paper Award HotStorage '24
- 3× SIGMOD 2024 · 2025 · 2026
- Google PhD Fellowship Nominee ASU nominee, 2026
What I Work On
the work →Key-Value Stores under Disaggregation
What breaks when an LSM engine's storage moves away from its compute, and can it stay fast, tunable, and encrypted once it does?
- 3× SIGMOD
- CaaS-LSM · SHIELD · O3-LSM
- Encryption & offloading
Filesystems and Distributed Caches
How do you hold a globally distributed, client-side cache coherent, and what is the system underneath actually costing you?
- CacheFS @ HPE Labs
- FUSE · RDMA · eBPF
- 2 patents under review
Highlighted Publication
all publications →Recent news
all news →-
“Probe Less, Learn More: Characterizing and Correcting eBPF Instrumentation Overhead” was accepted to the ProTools 2026 workshop! 📄
-
Our paper “TellyTune: Exploring the Limits of LLM-Driven Tuning for LSM-Based Key-Value Stores” was accepted to SYSTOR 2026! 📄
-
Received Invitation to attend the HPTS 2026 Conference! 🎉
-
Presented “O3-LSM” and “ANCHOR” at ACM SIGMOD 2026! 🚀 + 🔐
-
Honored to be ASU’s nominee for the Google PhD Fellowship Award! 🎉