Pandelis Zembashis

A benchmark for long-horizon organizational memory in AI agents: a synthetic, bi-temporal, multi-author/multi-channel corpus with a evaluation harness over publicly available memory systems.

Garry's Opinionated OpenClaw/Hermes Agent Brain

Verified RL dataset + native-toolchain graders for training models to edit zerolang projects (Prime Intellect / Laguna XS.2). 126 before/after tasks.

Ten medium-complexity CLI tools written in zerolang (grep, wc, bc, sort, cut, tr, base, factor, uniq, stats) — a corpus for LLM code-editing experiments.