About Oraculum
A small lab. A wider curiosity.
Oraculum, LLC is an independent AI research lab. We study how models make decisions and how neural signals can become useful interfaces, and we publish what we build as open models with their evidence attached.
What we believe
Intelligence is interesting. Making it useful is the work.
We care about the moment a signal becomes a representation, and the moment a model’s output becomes a decision someone can act on.
That takes more than a model. It takes a clear contract, careful evaluation, respect for uncertainty, and an honest account of what has not been shown yet.
↳How we release
Four habits, shown in public.
01Seal the test.V3 was fixed before its sealed unseen-task and computed-family evaluations were opened; gains and regressions are both reported.02Gate the port.Core ML conversions are checked against PyTorch before they ship: 20-step diffusion parity for every ZUNA profile, 210 of 210 real trials for motor imagery.03Pin the bytes.manchego-serve checks the SHA-256 of every weight file and reports, in every response, whether it loaded the published ones.04Lead with the weaknesses.The V3 card names its task-transfer uncertainty, unfamiliar-task overconfidence, and out-of-distribution regressions.
- Lab
- Oraculum, LLC
- Publishes as
- oraculumai on Hugging Face
- Open source
- manchego-serve, and a proposed numerical fix to coremltools
- Releases
- 7 in the catalog, 3 decision-model formats and 4 EEG packages
Find the lab
Keep the conversation open.
Explore releases, read their model cards, and join the technical discussion on Hugging Face. If a comparison ran your model wrong, the model card’s discussions are the place to say so.
Oraculum on Hugging Face