https://jackie-cqz.github.io/Notes on agent memory, developer tools, privacy computing, and motion control.enSat, 17 Oct 2026 12:00:00 GMThttps://jackie-cqz.github.io/posts/memory-across-sessions/https://jackie-cqz.github.io/posts/memory-across-sessions/Sat, 17 Oct 2026 12:00:00 GMTAgent memoryStarting with two automatically generated memories of AgentMeld's working conventions, I walk through prompt capture, memory recall, and work handoffs in my ZCode integration with PowerContext.https://jackie-cqz.github.io/posts/encrypted-recurrent-navigation/https://jackie-cqz.github.io/posts/encrypted-recurrent-navigation/Thu, 15 Oct 2026 12:00:00 GMTPrivacy computingAn experiment in continuous encrypted navigation: CKKS, a polynomial GRU, persistent ciphertext state, and a 2 Hz closed loop.https://jackie-cqz.github.io/posts/jev-in-deepseek-harness/https://jackie-cqz.github.io/posts/jev-in-deepseek-harness/Tue, 22 Sep 2026 12:00:00 GMTDeveloper toolsWhat caught my attention about TypeSafe Jev, and how I connected its small, typed decisions to DeepSeek Harness.https://jackie-cqz.github.io/posts/agentmeld-evidence-driven-context/https://jackie-cqz.github.io/posts/agentmeld-evidence-driven-context/Fri, 03 Jul 2026 12:00:00 GMTAgent memoryEvidence-driven context in AgentMeld: selecting active facts, upstream artifacts, and failure evidence so each attempt knows its task, its sources, and how to continue.https://jackie-cqz.github.io/posts/vlm-vla-manipulation/https://jackie-cqz.github.io/posts/vlm-vla-manipulation/Mon, 01 Jun 2026 12:00:00 GMTMotion controlConnecting Qwen2.5-VL task understanding, π0 action generation, and geometric verification in a layered Franka Panda simulation.https://jackie-cqz.github.io/posts/wheeled-leg-ppo-control/https://jackie-cqz.github.io/posts/wheeled-leg-ppo-control/Fri, 20 Mar 2026 12:00:00 GMTMotion controlFollowing a wheeled-leg robot's PPO control pipeline through observations, rewards, mixed action semantics, and deployment from Isaac Gym to ONNX and ROS.