AI & ML interests: Open reasoning models for local deployment. Post-training, agentic tool use, psychometrics-grounded evaluation, and efficient inference (GGUF, MLX, NVFP4).
Gestalt Labs 🇨🇦
Independent Canadian open reasoning research
Gestalt Labs builds and releases fully permissive open-weight reasoning
systems: multimodal assistants, tool-using agents, and local-first
deployments. Our methods are grounded in psychophysics and measurement
science, applying signal detection theory, adaptive staircase procedures,
and psychometric modeling to post-training and evaluation.
Models ·
Datasets ·
GitHub
Methods
- PEST — Per-Expert Staircase Thresholding. Psychophysical adaptive
staircase procedures (Taylor & Creelman, 1967) applied to MoE expert
pruning.
- NSC-ACE — contrastive steering directions extracted from hidden
states, refined with GRPO rollouts.
- Curated data pipelines — DDM-inspired curation achieving AUC 0.97
with 53% token savings at 99.5% sensitivity.
Model lines
Ornstein — multimodal and MoE reasoning models (27B–35B), shipped in
GGUF, MLX, and full-precision formats for every common local stack.
Harmonic — compact reasoning models with structural output
supervision (100% structural gate pass rate at 9B).
Architectures
- MPKnet — biologically inspired
CNN modeling the M/P/K pathways of the LGN. Competitive accuracy with
52× fewer parameters than ResNet18; runs on a Raspberry Pi.
Patent pending.
Featured datasets
Principles
- Ship open artifacts people can inspect, run, and adapt without restriction
- Pair every release with practical local formats (GGUF, MLX)
- Treat datasets as first-class research objects
- Measure everything: pruning ships with psychometric thresholds
Contact
research@gestaltlabs.ca · or open a discussion on any repository
