Releases and publications
Newest first. Each entry says what it is and how its numbers were obtained: measured by us, reproduced by us, or reported by others.
Primus
Primus for invoice decisions
26/32 structured reconciliation answers in the broader synthetic study, with all task conditions, competitor results and public evidence.
Read the case studyPrimus Decision 0.1 — Research Alpha
Primus records 75.1% accuracy, NLL 0.637 and raw ECE 0.127 with 3,715,074 neural parameters. AAME's reproduction of the 421M-parameter Laya checkpoint on the same harness records 76.6% accuracy, NLL 0.707 and ECE 0.213. Primus uses about 113× fewer neural parameters; its separate feature tables bring the installed release to 157.9 MB. Full task results and measurement conditions accompany the comparison. M, measured by us R, reproduced by us
Introducing Primus Decision 0.1
Why not every AI action should be generated text, what typed probabilistic decisions are, how the model was evaluated, and the research ahead.
Read the articleThe Road to Primus
One public decision component within a programme connecting memory, reasoning, learning and action. Public evidence and protected implementation have a clear boundary.
Earlier papers on Zenodo
Earlier AAME preprints document graph-based research preceding this trained decision-model release. Each paper retains its own methods and claims; Decision 0.1's results are documented separately. Titles, dates and DOI links preserve that research record.
- Preprint2026-02-27
Continuous Memory Consolidation Without Retraining: Solving the Edge Bloat Problem Through Computational Sleep
Pally, Sai Tilak · doi:10.5281/zenodo.18801534
Preprint · Zenodo · open access
- Preprint2026-02-26
Causal Imagination Without Language Generation: Forward Simulation Through Graph Traversal
Pally, Sai Tilak · doi:10.5281/zenodo.18779870
Preprint · Zenodo · open access
- Preprint2026-02-25
Primus: Cognition From First Principles — The Vyan Dual-Graph Cognitive Architecture Without Language Models
Pally, Sai Tilak · doi:10.5281/zenodo.18739227
Preprint · Zenodo · open access
This paper documents earlier graph-based Primus research. The trained Decision 0.1 component has its own release and evaluation record.
- Preprint2026-02-22
Cross-Domain Transfer Learning in Developmental Knowledge Graphs Without Training
Pally, Sai Tilak · doi:10.5281/zenodo.18734945
Preprint · Zenodo · open access