Rahul Anand Sharma

I am an ML Systems Engineer & Researcher at Google Cloud, specializing in hardware-software co-design, distributed accelerator systems, and high-performance inference.

At Google Cloud, I focus on inference optimization of Gemini models across GPUs and TPUs, and previously drove accelerator enablement and scaling analyses for TPU v7 (Ironwood) (Google Cloud Tech Impact Award, 2026).

Earlier, I was an Applied Scientist at AWS architecting GenAI root-cause analysis for Apache Spark at hyperscale (Amazon Q), completed my Ph.D. in Electrical & Computer Engineering at Carnegie Mellon University (CyLab, CIT Dean’s Fellow), and conducted systems research at Microsoft Research (FarmBeats, DeepEdge).

My work spans 15+ publications in top systems and security venues (SIGCOMM, NSDI, USENIX Security), 2 US patents, and systems recognized by GatesNotes, Satya Nadella, and the front page of Hacker News.

---

Core Focus

• LLM Inference & Disaggregated Serving: Gemini model serving across GPUs and TPUs, intra-node prefill/decode disaggregation, KV-cache optimization, model parallelism, and low-latency runtimes.
• Accelerator Infrastructure & Scaling: TPU v7 (Ironwood) multi-cube scaling, collective communication (AllReduce/AllGather/AllToAll), XLA/JAX runtime optimization, and interconnect bottleneck modeling.
• Secure AI Infrastructure & Network Security: Air-gapped / Sovereign Cloud LLM serving with strict memory and control-plane isolation (Google Distributed Cloud), hyperscale active network security validation (AWS), and ML-driven network anomaly detection (CMU CyLab, USENIX Security).

---

Selected Production AI Systems