Building data platforms and AI systems.
Google Cloud • Southeast Asia
- AI Interpretability: Activation probing, steering vectors, sandbagging detection
- Distributed Systems: Formal verification, UPIR framework, large-scale data platforms
- LLM Infrastructure: Evaluation at scale, context management (MCP), agentic architectures
- Beating CUDA with Triton: A Fused MoE Dispatch Kernel for Mixtral and DeepSeek - Pure Triton MoE dispatch that beats Megablocks at inference batch sizes
- Attention Is All You Bid - Transformer architectures for real-time bidding
- Three Bets on Model Honesty - Where AI alignment research is heading next
- Circuit Tracing in Production - Mechanistic interpretability for production AI safety
- From 11% to 88% Peak Bandwidth: Custom Triton Kernels for LLM Inference - The foundation for the fused MoE work
→ More at subhadipmitra.com/blog
Data Platforms │ AI/ML Infrastructure │ Research
───────────────────────┼──────────────────────────┼─────────────────────
Context-Aware Pipelines│ LLM Evaluation │ Interpretability
ETLC / Context Stores │ Model Serving │ Activation Steering
Agent-Ready Platforms │ MLOps / LLMOps │ Formal Verification
Dynamic Context Engines│ Agentic Systems │ AI Safety
Singapore • Google Cloud • Schedule a call