I build research systems that can run for a long time without becoming a black box.
Research should be executable: observable, testable, and recoverable.
X / SYSTEMSAgentic R&D systems Long-horizon agents, evaluator gates, multi-agent harnesses, recovery protocols, and token economics. |
Y / AUTONOMYAutonomous flight & RL Fixed-wing control, visual localization, sensor fusion, simulation, and reinforcement-learning navigation. |
Z / EARTHGeospatial vision & data Remote-sensing pipelines, multimodal perception, dataset generation, and map-to-model tools. |
token-firewall-teamDelegates implementation to lower-cost workers and reserves frontier-model tokens for bounded review, with reproducible evaluation evidence. |
autoresearch-paperRuns long-horizon paper projects with frozen evaluators, research gates, watchdogs, and resumable state. |
daily-cockpitA local-first Obsidian workspace that protects today, captures ideas without pressure, and reconnects yesterday's context. |
codeinRuns Codex from an Obsidian sidebar through the local CLI, with explicit model, reasoning, and file-access controls. |
pilot_rl_navigationExplores drone navigation through visual positioning, sensor fusion, reinforcement learning, and recovery control. |
RSDatasetGeneratorTurns geospatial vector points into remote-sensing image datasets with tiled downloads and pixel-coordinate metadata. |
Long-horizon agent evaluation, UAV agent harnesses, and knowledge systems that maintain themselves.
LLM-Skills · ztrade-ares-7x24 research sandbox
Field notes · ORCID