Deep-tech research lab
We build Thin Language Models from the ground up: data, distillation, and weights that ship.
Move your cursor through the data
A Thin Language Model is small, specialized, and runs locally: frontier-grade judgment for one domain, distilled into weights that fit on a laptop. These are the models we research and release.
A Thin Language Model for the factory floor. Grounded reasoning across operations, quality, and supply chain, running locally on Apple Silicon.
A Thin Language Model for law. Runs standalone, or as a Lens in front of a frontier model, scoping jurisdiction, citations, and routing before the expensive call.
A Thin Language Model for growth. A Gemma 2 9B lens for SEO and paid-platform marketing: audits, Search Console and Google Ads framing, on-page and backlink analysis, RAG-grounded, with deterministic guardrails that refuse black-hat tactics by design.
Need a small language model instead? We build bespoke SLMs as a service: your domain, your data, your hardware. The TLMs above are what the lab ships under its own name.
Commission an SLM →Three tiers, one trade-off. Generality is broad, hosted, and costly. Specialization is narrow, local, and cheap. We engineer the right tail.
Our agenda runs from data infrastructure to emergent model behaviour. Every thread feeds the models we ship.
Proprietary pipelines that surface signal from noise at scale, semantically indexing billions of data points.
The category we coined: frontier-grade reasoning distilled into sub-2B models that run locally and outperform far larger baselines on their domain.
Multi-step inference engines that decompose novel problems, form hypotheses, and self-correct in real time.
Pipelines producing training data indistinguishable from real-world distributions, closing the data scarcity gap.
Mapping the boundary between programmed behavior and spontaneous cooperation to exploit emergent dynamics.
10x size reduction with under 3% accuracy degradation, with large-scale capability on constrained hardware.
AI-powered prospecting, pipeline, and relationship intelligence. Turn signals into pipeline, automate outreach at scale, and close smarter deals faster.
Visit Website ↗Venture intelligence for smarter fundraising and introductions. Data-driven dealflow, the right investors, and networks that compound.
Visit Website ↗One philosophy, drawn in two colors: what we refuse, and what we own.
Open at the foundation. Ours at the edge.
Not the volume. The method. We build training data the way we build models: in-house, synthetic where it must be, and reviewed by people who know the domain.
Reach the lab directly.
We collaborate with researchers, enterprises, and builders who are serious about AI. If you are building something that matters, we want to hear from you.
Start a conversation →