About Me
Rabimba Karanjai is a Staff Agentic AI Researcher at PayPal Research. He completed his Ph.D. in Computer Science at the University of Houston (defended July 2025), where his research focused on the intersection of blockchain, secure AI reasoning, and LLM-assisted program analysis and repair.
Education
Ph.D. in Computer Science
University of Houston
M.S. in Computer Science
Rice University
Interests
Agentic AI & AI Reasoning
LLMs Security & Smart Contracts
Decentralized Systems & Blockchain
Formal Verification & Software Engineering
Web VR & High Performance Computing
Research Overview
I am a Staff Agent Researcher at PayPal and a PhD graduate from the University of Houston, advised by Prof. Weidong (Larry) Shi. My research sits at the intersection of AI reasoning, decentralized systems, and software security.
Core research areas:
- LLM Reasoning & Code Generation — Multi-agent architectures for reliable code/test generation (AgentTester, LogBabylon) and hallucination mitigation
- Blockchain & Smart Contract Security — Cross-chain vulnerability detection, decentralized infrastructure, and formal verification of smart contracts
- Quantum-Enhanced ML — Quantum contrastive embeddings for near-term devices (QuCoWE, AAAI 2026)
I am a Google Developer Expert in Web Technologies & ML, a 3x Sui Foundation Research Award recipient, and serve as Associate Chair for CSCW 2026.
Key Achievements
37
Peer-Reviewed Publications
28 as first author across USENIX Security, AAAI, ACM FSE, MICRO, IEEE ICBC, KDD — 10 more under review
1
Issued US Patent
US11025643B2 — multi-factor authentication combining biometrics and blockchain (IBM)
4x
Research Grants
$90K+ from Sui Foundation (3x) and Grant for Web as Principal Investigator
85+
Invited Talks
Google I/O, OSCON, NDC Sydney, FOSDEM, KubeCon, Stanford, MIT, Berkeley RDI, and more
ACM SIGSOFT Distinguished Paper Award
ACM AIWare 2025 — Securing the Multi-Chain Ecosystem
Best Dissertation Award
Dan E. Wells Outstanding Dissertation Award & UH CS Best Dissertation (2025)
Google Developer Expert
Web Technologies (2018–present) & Google Cloud AI/ML (2024–present)
NVIDIA Developer Champion
Recognized for contributions to AI and GPU computing community (2026)
Distinguished Publications
Collaboration is all you need: LLM Assisted Safe Code Translation
Rabimba Karanjai
Smart Contract Code Translation based on Concepts
Rabimba Karanjai
Decentralized Application Infrastructures as Smart Contract Codes
Rabimba Karanjai
Grants & Funding
- Sui Foundation Academic Research Award (2023, 2024, 2025) — Primary Investigator, $25,000/year for research on decentralized infrastructure and smart contract security
- Grant for Web — Primary Investigator, $15,000 for advancing privacy-preserving and decentralized web technologies
- Google Cloud Research Innovator (2023, 2024) — Cloud credits and research support for AI/ML projects
Research Projects
Bay Area Pujo Parikrama
Interactive Durga Puja guide and route planner for the San Francisco Bay Area, mapping 31 pujas across 6 regions with schedules, bhog details, tickets, and directions.
•
1 min read
Houston Pujo Parikrama
Community guide and interactive pandal-hopping planner for Durga Puja across Greater Houston, covering schedules, bhog, and cultural events across 5 regions.
•
1 min read
AI Race Coach (Project Koru)
Built a real-time AI Race Coach at Sonoma Raceway with Google Developer Experts, using Antigravity, Gemini API, Gemma 4, and Jetpack Compose to process live telemetry at 100 mph.
•
1 min read
ConferenceRank
One-stop shop to evaluate CS conference & journal venues — CORE/ICORE ranks, SJR quartiles, multi-decade acceptance rates, submission deadlines (AoE), and an MCP server + JSON API …
•
1 min read
Open Source
One-stop CS venue evaluation — CORE/SJR ranks, acceptance-rate trends, AoE deadlines, and venue suggester. Ships MCP server, JSON API, and agent skill for LLM consumption.
Privacy-first DICOM viewer that plans slice selection with Gemma 4 before multimodal review — ~99% fewer image tokens, fully in-browser via WebGPU.
Single-variable benchmark harness for serving LLMs with vLLM across CUDA and OpenXLA/XLA backends — honest goodput under fixed SLOs.
The AI Race Coach Android dashboard from the Google I/O GDE Field Test — 10 Hz telemetry, Gemini reasoning, on-device Gemma 4 coaching.
Artifacts for our USENIX Security 26 paper: the LogInject benchmark, Context Stitching attacks, and layered-defense mitigation evaluation.
Co-authored Google codelab “Building Trustable AI at 100 MPH” — split-brain architecture, telemetry streaming, guardrails, and Gemini strategy layers.
Research prototypes, agents, and tooling — from smart-contract security scanners to multi-agent routers.
Google Developer Expert & Community
- Google Developer Expert in AI/ML & Cloud (previously Chrome & Web, 2018–2024) — directory profile
- Co-author of Google’s official Building Trustable AI at 100 MPH codelab with a global GDE cohort
- Invited speaker at Google I/O Field Tests (Sonoma Raceway 2026), OpenXLA DevLab Summer 2026, and the GDE Summit at Google I/O 2025
- Google Cloud Research Innovator (2023) and Google Cloud Champions Innovator (2023–2025)
- Mentor & judge for the Google Solutions Challenge and Women Techmakers Academy; earlier Mozilla Tech Speaker and Mozilla Research Fellowship (2018)
Recent Publications
(2026).
Specialists Hold, Generalists Discount: Asymmetric Equilibrium in LLM Routing Auctions.
NeurIPS 2026.
(2026).
Reachability-Based Capability Confinement for LLM Agents under Indirect Prompt Injection.
arXiv:2608.30041.
(2026).
PACE: Policy-Attested Contract Execution for Safe AI Agents in Decentralized Finance.
BRAINS 2026.
(2026).
When Agents Act on Web3: An Attack-Surface Survey of MCP, Skills, and Tool Calling.
BRAINS 2026.
(2026).
Allocation Before Ranking: Decoupled Token Compression for OmniLLMs.
NeurIPS 2026 (Under Review).
Upcoming Talks
Recent Talks
House of AI: AI Safety Forum — Closing Panel
Closing panel at the House of AI Safety Forum (0G Labs, Stanford Founders Club) on *The $100 Billion Question: Who Is Responsible When AI Goes Wrong?* — with Jon Ozdoruk (DSALTA) …
Context Contamination at USENIX Security 26 — Oral Presentation & Poster
Presented our paper "Context Contamination in LLM Analysis of Network Security Logs" at USENIX Security 2026 in Baltimore — an oral presentation plus a poster session.
OpenXLA Summer DevLab 2026 — Benchmarking vLLM on XLA Backends
Invited talk at Google OpenXLA DevLab Summer 2026 on "Benchmarking vLLM on XLA Backends: From CUDA to OpenXLA for LLM Serving", speaking alongside engineers from Google, NVIDIA, …
Google I/O GDE Field Test — AI Race Coach at Sonoma Raceway
Built a real-time AI Race Coach with Google Developer Experts using Antigravity, Gemini API, Gemma 4, and Jetpack Compose. Processed live telemetry at 100 mph to deliver …
Beyond the Chatbot: A Blueprint for Trustable AI
Talk at Google DeepMind on building trustable AI systems beyond simple chatbots. **Key points:** - Moving from conversational AI to agentic AI architecture - Safety frameworks for …
Recent Blog Posts
Racecraft: Fine-Tuning Gemma 4 with DPO for Sub-150ms On-Device Coaching at 130 MPH
Read full post →
Splitting the Brain to Beat the Clock
→ paste between BEGIN END markers.
The Throughput Trap: Benchmarking vLLM on OpenXLA and the Reality of Production LLM Serving
vLLM Systems · DevLab 2026, Deep Dive I was recently invited by the Google TPU team to speak at the OpenXLA Summer DevLab 2026 . This post breaks down our deep-dive evaluation of …
Professional Service
- Associate Chair, ACM CSCW 2026
- Scientific Advisory Committee, Texas Quantum Initiative (2025)
- Google Developer Expert, Web Technologies (2018-present) & Google Cloud (2024-present)
- Reviewer: IEEE ICBC, ACM CSCW, BRAINS
Get in Touch
Happy to talk about research collaborations, industry problems, talks, or the GDE program. The fastest way to reach me is email or LinkedIn.