CURSED HELM is an independent research lab from Aotearoa New Zealand founded by Fraser Austin. We craft bespoke frameworks for AI autonomy and advise on AI‑driven information sovereignty.
Curious Projects:
defaultMODE — Python framework for Discord‑native AI entities with persistent memory, selective attention, and reflection. Supports multi‑agent teams, ensures long‑run coherence, and provides full SQL / JSONL audit logging.
PerSIMS — needs‑driven life‑sim where an embodied LLM agent juggles hunger, hygiene, energy, and economy, revealing emergent behaviour under resource constraints.
REPTARindex — recursive embedding + inverted‑index retriever that builds hierarchical clusters and summaries for rapid, brain‑like recall across large text troves.
LLMdiscoEVAL — playful evaluation harness where language models perform collaborative “dance routines,” scored across rounds to measure synergy and creativity.
MoRe — “Mixture of Rankers” ensemble judge that merges multiple LLM‑as‑judge scorers into a balanced consensus quality metric.
qBERT — uses off the shelf BERT models as a language model, using coherence and memory.