At ZkAGI, I built HAISA, a privacy-preserving LLM inference compiler that splits transformer linear operations across miners while keeping non-linear operations on-client. It supports HuggingFace models including Llama, Qwen, and Mistral, with information-theoretic privacy and bit-identical output.
I also designed and shipped PawPad, a privacy-preserving wallet and autonomous trading system on Oasis Sapphire ROFL, live on mainnet with EVM and Solana support. For ZkTerminal, I built backend APIs, async signal pipelines, and DEX integrations, and reduced end-to-end inference latency by 25% through pipeline redesign and request batching.
At ZkAGI, I productionized ML and LLM workloads across AWS SageMaker, Bedrock, GCP Confidential Containers, and on-prem GPUs. Before that, at Fintricity, I developed backend services and data pipelines for sustainability software, including carbon-footprint calculations and auditable reporting APIs.
