LLM Safety & Adversarial Testing
ActiveQuality-diversity evolutionary framework for discovering diverse vulnerabilities in LLMs. Accepted at ICLR 2026 Workshop AIWILD.
What I'm working on, learning, and thinking about right now. Inspired by nownownow.com.
Open to Consulting Singapore
Last updated: 2026-03
Current Projects
Orbital Compute Primitives
ActiveDesigning compute abstractions for environments where power, connectivity, and time are all intermittent. Scheduling and fault-tolerance driven by orbital physics.
AI Agent Framework for Data Engineering
ActiveBuilding production-ready multi-agent systems for autonomous data pipeline management. Industry-agnostic framework handling data quality, orchestration, and monitoring.
LLM Inference Efficiency Research
ResearchReference implementations of acceleration techniques: speculative decoding, KV-cache compression, custom Triton kernels. 8.1x speedup, 88% peak bandwidth on A100.
Circuit Tracing & Interpretability
ResearchMechanistic interpretability research for production AI safety. Circuit tracing techniques for understanding model internals and sandbagging detection.
5
Active Projects
7
Countries This Year
8
Technical Disclosures
2
New Papers
Learning & Research
- Multi-Agent Orchestration: Coordination protocols and fault tolerance for production systems
- Orbital Machine Learning: ML for orbit determination, conjunction forecasting, and satellite behavior classification from sparse observations
- Distributed Computing for Space: Gradient compression for high-latency links, model partitioning across heterogeneous nodes, and synchronization under intermittent connectivity
Currently Reading
Low Earth Orbit Satellite Design
George Sebestyen - Satellite systems architecture and distributed networks
Why Machines Learn
Anil Ananthaswamy - Mathematical foundations of machine learning
What I'm Not Doing
- Not taking on new full-time commitments (happy with current role)
- Not doing basic web development projects (focused on AI/data architecture)
- Not available for unpaid advisory work (limited capacity for pro-bono)
Open For
- Technical Consulting: AI/data architecture, framework design, team building
- Speaking Engagements: Conferences, workshops, podcasts on AI/data topics
- Research Collaboration: Academic partnerships on agentic systems or context modeling
- Open Source: Collaborators for my frameworks and tools