Projects
Research software, evaluation frameworks and reference implementations. Only projects explicitly configured as Institute outputs or collaborations appear here — Octopus Core commercial products are not listed as Institute research.
- Status: ExploringResearch softwarePose & Landmark Evaluation (Auslan exploration)
Exploratory tooling to evaluate pose/landmark extraction quality for multimodal Auslan research. No trained model or dataset is released.
- Status: Active researchEvaluation frameworkModel Evaluation Harness
A framework for reproducible model evaluations that records dataset/model versions, seeds, hardware and metric definitions alongside results.
- Status: Active researchResearch softwareoctoryn-router (sovereign DP router)
A hand-written sovereign HTTP reverse-proxy router over a data-parallel inference fleet, with mid-stream byte-identical failover and an event-logged audit trail.
- Status: Active researchEvaluation frameworkrocm-bench (HF-logits parity harness)
A logits-parity harness that runs a model and dumps per-token logits for offline comparison against a HuggingFace oracle on identical token ids. Does not use vendor BLAS.
- Status: Active researchResearch softwaremac-lock (cross-track RAM mutex)
A vm_stat-gated, mkdir-atomic mutex that serialises memory-heavy work across two tracks on a single 128GB Mac, with a 40GB headroom self-check and stale-lock auto-reclaim.
- Status: ExperimentalReference implementationGoverned Execution Boundary (reference implementation)
A reference implementation of a single policy-enforcing execution boundary with human approval and evidence capture.