Rabimba Karanjai
  • About
  • Publications
  • Projects
  • Talks
  • Blog
  • Experience
  • CV
ESC

Searching...

No results found

↑↓ Navigate ↵ Select
Powered by Hugo Blox
  • Publications
    • Empowering Smart Contracts with Real-time On-Chain AI Inferences
    • Rethinking Publication: A Certification Framework for AI-Enabled Research
    • CausalGraphX: A Counterfactual Graph Neural Network Framework for Explainable Systemic Risk Assessment
    • QuCoWE: Quantum Contrastive Word Embeddings with Variational Circuits for Near-Term Quantum Devices
    • Aligning Vision Language Models via anchor
    • Allocation Before Ranking: Decoupled Token Compression for OmniLLMs
    • Bridging Confidentiality and Reliability: Open-Weight Agents for Legal Reasoning
    • Closing the Cost Gap in Verifier-Guided RL: A Smart Contract Case Study
    • CM-DPO: Constraint-Margin Direct Preference Optimization for LLM Planning
    • Hype or Hope? Training LLMs on Decentralized GPU Clouds
    • Latency-Optimized Cascaded Voice Agent Pipeline
    • Profiler: Per-Question Persona Conditioning of Frozen LLMs via Retrieval and Fusion of Synthesized Profiles
    • Rules That Govern: A Per-Property Analysis of Multi-Agent LLM Deliberation
    • Severity-Aware Preference Training Framework for Constraint-Satisfying AI Planners with Minimal-Edit Supervision
    • Specialists Hold, Generalists Discount: Asymmetric Equilibrium in LLM Routing Auctions
    • Thread-Based Conversational Commerce System with Multi-Modal Ordering and Real-Time Fulfillment Tracking in Messaging Platforms
    • Unlocking On-Chain Intelligence: A Practical Framework for GenAI-Powered Smart Contracts
    • VerifyGen-X: Secure Cross-Chain Smart Contract Generation via Scalable Reinforcement Learning from Formal Verification Feedback
    • HPCAgentTester: A Multi-Agent LLM Approach for Enhanced HPC Unit Test Generation
    • Securing the Multi-Chain Ecosystem: A Unified, Agent-Based Framework for Vulnerability Repair in Solidity and Move
    • Unlocking On-Chain Intelligence: A Practical Framework for GenAI-Powered Smart Contracts
    • Evaluating the Quality of Randomness and Entropy in Tasks Supported by Large Language Models
    • Teaching AIs to Reason and Code, Confidentially
    • Collaboration is all you need: LLM Assisted Safe Code Translation
    • Ransomware 3.0: Enhancing Risk Management and Mitigation Options with Proof-of-Decryptability and Smart Contracts
    • Synthesizing Public Opinions with LLMs: Role Creation, Impacts, and the Future to eDemocracy
    • LogBabylon: A Unified Framework for Cross-Log File Integration and Analysis
    • Artificial Intelligence-Based Target for Personalized Interventions of Atherosclerosis from Gut Microbiota Signature
    • Smart Contract Code Translation based on Concepts
    • TPU as Cryptographic Accelerator
    • Adding All Flavors: A Hybrid Random Number Generator for dApps and Web3
    • Enhancing Vascular Disease Diagnosis through AI-Driven Analysis of Histopathology Images
    • Evaluating Irrationality in Large Language Models and Open Research Questions
    • Mitigating Hallucinations in AI-Driven Medical Diagnosis
    • Trusted LLM Inference on the Edge with Smart Contracts
    • All We Need is Voter Feedback
    • Decentralized FaaS over Multi-Clouds
    • DIaC: Re-imagining Decentralized Infrastructure As Code using Blockchain
    • Who is Smarter? An Empirical Study of AI-based Smart Contract Creation
    • Decentralized Machine Learning Governance: Overview, Opportunities, and Research Challenges
    • DeFaaS: Decentralized Function-as-a-Service for Emerging dApps and Web
    • DHTee: Decentralized Infrastructure for Heterogeneous TEE
    • Supporting Heterogeneous TEE for Critical Infrastructure Protection
    • Decentralized Application Infrastructures as Smart Contract Codes
    • Privacy preserving event based transaction system in a decentralized environment
    • Lessons Learned from Blockchain Applications of Trusted Execution Environments and Implications
    • An event driven framework for smart contract execution
    • Mobile multi-party digitally signed documents and techniques for using these allowing detection of tamper
    • On Conditional Cryptocurrency With Privacy
    • Optimizing Web Virtual Reality
  • Blog
    • Splitting the Brain to Beat the Clock
    • The Throughput Trap: Benchmarking vLLM on OpenXLA and the Reality of Production LLM Serving
    • Teaching the Coach to Read the Driver
    • Racecraft (Project Koru) · Prologue — The Origin Story
    • The 800-Millisecond Problem
    • The Blind Spot Horizon: Why Your AI Benchmarks Are Lying to You
    • My Friend's MRI Didn't Come with a Manual, So I Built One with AI
    • The Hidden Performance Trap in Causal Ring Attention
    • Unboxing Gemma: High-Throughput Sparse Autoencoder (SAE) Training on Google Cloud TPU v5e
    • Building a Multimodal Document Intelligence Pipeline with Gemma 4 E2B
    • The Gemma 4 E2B Fine-Tuning Cookbook
    • Write Once, Scale Everywhere
    • Visualizing the Invisible: How Nano Banana 2 Turns Dense Science into Stunning Art
    • A Split‑Brain Neuro‑Symbolic Training Method for High‑Velocity Autonomous Coaching from Telemetry
    • Fine-Tuning at the Redline: Architecting a Trustable AI Racing Coach
    • The Context Trap: Why Scaling Laws Can’t Break the Ceiling of Uncertainty
    • Building Production AI Systems with Free Cloud GPUs
    • 🎤 SpeakWise: Build an AI Public Speaking Coach with Gemma 3n
    • From Models to Agents: Shipping Enterprise AI Faster with Google’s MCP Toolbox & Agent Development Kit
    • Build Smarter AI Agents Faster: Introducing the Google Agent Development Kit (ADK)
    • Deep Dive into the Google Agent Development Kit (ADK): Features and Code Examples
    • My Google I/O 2024 Adventure: A GDE's Front-Row Seat to the Gemini Era
    • MovieBuff: Dive Deeper into Movies with Generative AI
    • Curious case of Cisco AnyConnect and WSL2
    • Immersive Payment: Second Update
    • Immersive Payment: First Update
    • Road to the Grant : Immersive Payment
    • Attack of the Clones (Part 1): Drishti and vMeet
    • Handling Web Server file permission: The Last Guide
    • Help finding a cure for COVID19: Set up Folding@Home
    • Blocking Communication - A perspective from India : The Beginning
    • Blocking Communication - Breaking Out
    • Visualizing large scale Uber Movement Data
  • Curriculum Vitae
  • Talks & Presentations
    • Beyond the Chatbot: A Blueprint for Trustable AI
    • United in Defense: Architecting Safe and Trustworthy AI Agents
    • Google I/O 2025: GDE Summit, From Whiteboard to Users
    • From Whiteboard to Users: Making Research Accessible
    • Unmasking the Shadows: AI Red Teaming in the Age of Gemini and VertexAI, fortified by SAIF
    • Supercharging GenAI: Ray, Kubernetes, and TPUs for Lightning-Fast Inference
    • LLM, Reasoning and Agentic Gemma
    • LLMinABox: On Device Personalized Diary & Concierge using your voice and Gemma
    • AI, LLMs, and the Future of the Web: A Conversation with Rabimba and Hemanth (PayPal) at Google I/O
    • Berkley RDI: Code Translation
    • Build with AI Intro to Function Calling with Gemini by Rabimba Karanjai GDE 2024 03 20 18 07 GMT 5
    • Building Decentralized Social Virtual Reality using WebXR on your browser
    • Collaboration Is All You Need: LLM Assisted Safe Code Translation
    • GDG Houston Presents: IO Extended Web Edition. The state of WebXR by Rambimba Karanjai
    • GDG Zagreb - Online Tech Talk #1
    • GDG Zagreb - Online Tech Talk #1 - State of WebXR and building with AR/VR today - Rabimba Karanjai
    • Google I⧸O Extended Edmonton 2024 Immersive art with AI
    • Hands-on Workshop: Develop an App with Vertex AI Gemini 1.0 Pro
    • LangChain Deep Dive with Rabimba Karanjai (Google Developer Expert)
    • Mozilla TS Paris Meetup Talk : Introducing jARvis
    • Rabimba Karanjai - A self learning word prediction module for handling multilingual input
    • SolMover: https://dl.acm.org/doi/10.1145/3664646.3664771
    • Stanford CS25: V2 I Emergent Abilities and Scaling in LLMs
    • State of WebVR, A-Frame and Our Foray Into Web Mixed Reality : Rabimba Karanjai - Mozilla
    • Talk to your Desktop, Code, PDF or anything, in any language, using your own LLM or Gemini
    • Testing Web VR/AR Mixed Reality Applications: What you need to know by Rabimba Karanjai #SeConf2020
    • This is what we mean when we say Google Developer Experts (GDE)
    • TPU As Cryptographic Accelerator: At Google Cloud Research Innovator Showcase
    • Voice Driven VR
    • Walkthrough of Gemini and Gemma on Cloud and On-device
    • Mitigating Hallucinations in AI-Driven Medical Diagnosis
    • SolMover: Smart Contract Code Translation Based on Concepts
    • Privacy aware Zero Knowledge Login using Oauth2 and Passkey
    • LLM Applications components and design patterns Hands on Workshop
    • Give your web apps superpower with Generative AI and Mediapipe
    • An Empirical Study of AI-based Smart Contract Creation
    • On Device Generative AI: Building your own Dall-E in the browser, welcome WebGPU
    • DeepSpeech: A Journey to <10% Word Error Rate
    • DeepSpeech Architecture at North Carolina DevFest 2022 at Chapel Hill
    • North America Connect 2022 ｜ Day 2 GDE Stage Rabimba
    • FOSDEM 2020 - Building Decentralized Social Virtual Reality using WebXR on your browser
    • State of WebXR: What do You Need to Know Today? - Rabimba Karanjai - NDC Sydney 2020
    • DevFest Ranchi 2019 – Rabimba Karanjai - DeepSpeech: A Journey to less than 10% Word Error Rate
    • Visualize your Data in a 3D VR world using A-Frame in WebVR
    • Turning sensors into signals
    • Hardening Your IoT Endpoints: A Preventive Toolkit
    • Optimizing Web Virtual Reality
    • Rabimba Karanjai: VR for Humans Build your world using WebVR and aframe | Mobile day 2017
    • SecurityPI: IronClad your Raspberry PI
    • State of WebVR & aframe: Yesterday, Today, and Tomorrow Beyond Horizon
    • OSB 2015 - Building a self learning word prediction and auto-correct module for FirefoxOS
    • QA Robotics at Confluence 2012
  • Projects
    • LLM Reasoning & Code Generation
    • Quantum Computing & AI
    • Blockchain & Decentralized Systems
    • Edge Computing & Trusted Execution
    • WebVR & Web Technologies
  • Experience

The Throughput Trap: Benchmarking vLLM on OpenXLA and the Reality of Production LLM Serving

Jun 25, 2026·
Rabimba Karanjai
Rabimba Karanjai
· 1 min read
Site

Read full post →

Last updated on Jun 25, 2026
AI Systems Benchmarking CUDA Google Developer Expe Google DevLab LLM Serving PayPal AI Lab TPU VLLM XLA
Rabimba Karanjai
Authors
Rabimba Karanjai (he/him)
Staff Agentic AI Researcher
← Splitting the Brain to Beat the Clock Jun 25, 2026
Teaching the Coach to Read the Driver Jun 15, 2026 →

© 2026 Rabimba Karanjai. This work is licensed under {license}