Output 18 papers, 8 packages
Latest ICLR 2026
Now New ventures
Singapore, 2026
About
Data platforms. AI systems. The infrastructure between them.
delta at Google Cloud — leading Data & AI innovation and transformation across Southeast Asia. Publishing research on multi-agent systems, inference optimization, and AI safety. Exploring what's next.
Now
What I'm up to →Building
ICLR 2026 paper on LLM safety. Compute primitives for orbital environments.
Exploring
New ventures at the frontier. Conversations with research labs and founders building what's next.
Writing
Notes on activation steering, circuit tracing in production, and bets on model honesty.
Research Focus
01
Multi-Agent Systems
Coordination protocols, debate frameworks, verification for enterprise AI
Read about Agent Protocol Stack →02
Inference Optimization
Speculative decoding, Triton kernels, computational efficiency at scale
Read about Triton Kernels →03
AI Safety
Mechanistic interpretability, circuit tracing, sandbagging detection
Read about Circuit Tracing →04
Distributed Systems
Formal synthesis, correct-by-construction architectures
Read about UPIR →Selected Publications
View all →- 2026
Spark-LLM-Eval: A Distributed Framework for Statistically Rigorous Large Language Model Evaluation
arXiv - 2026
Quality-Diversity Evolution for Discovering Diverse Vulnerabilities in LLM Safety
ICLR 2026 Workshop AIWILD - 2026
Field-Theoretic Memory for AI Agents: Continuous Dynamics for Context Preservation
arXiv - 2026
Automated Rule Generation for Tiered Systems Using Multi-Stage Failure Learning
Google, Technical Disclosure Commons
Recent Writing
View all →- Confessions vs. CoT Monitoring vs. Probes: Three Bets on Model HonestyMarch 2026
- OpenAI's Confessions Paper Has a Blind Spot. Here's What Fills It.February 2026
- Activation Steering in 2026: A Practitioner's Field GuideFebruary 2026
- Moltbook as MCP Stress Test: What 770K Agents Reveal About Protocol DesignFebruary 2026
- Circuit Tracing for the Rest of Us: From Probes to Attribution Graphs and What It Means for Production SafetyJanuary 2026
- RLVR Beyond Math and Code: The Verifier Problem Nobody Has SolvedJanuary 2026
Building something interesting? Let's talk.