Independent AI research and engineering

Useful AI systems, built from first principles.

YY is an independent AI research and engineering company. We build software today that people can run and verify for themselves, and we work toward research on models and compute over time. The through line is the same either way: understand the problem from the ground up, and ship something that holds up when it is checked.

What we work on

Two horizons, one method

We are a young company with a deliberately narrow surface area. What follows is what we are actually doing, not a roadmap of things we hope to announce.

  • Software now

    Small, self-contained tools that do one useful thing end to end, run where the work already happens, and can be inspected by the people who depend on them.

  • Models and compute over time

    Our longer-horizon work is research on models and the compute they run on. We would rather publish it when there is something concrete to show than describe it in advance.

  • First principles

    We start from the actual problem and the constraints around it — the data, the machine, the failure modes — instead of assembling whatever the current stack makes easy.

First public release

yy-corpus-evidence

Deterministic, content-addressed evidence for a local corpus.

Point it at a directory of files. It inventories the tree, hashes file contents, and writes a manifest that is byte-identical every time it sees the same supported inputs. Optional declarations record human assertions about the files and surface unknown, incomplete, or conflicting evidence. Later, the corpus and its declarations can be checked against the manifest for drift.

It reads your files locally. It does not upload corpus contents, and it does not draw legal or compliance conclusions on your behalf.

  • Python
  • Open source
  • Runs locally
  • Published JSON Schemas
Install
python -m pip install yy-corpus-evidence

How we build

Commitments that shape the work

  1. Determinism before convenience

    The same inputs should produce the same outputs, byte for byte. Results you cannot reproduce are anecdotes.

  2. Local by default

    A tool should be useful on one machine, offline, before it asks anyone to send data somewhere else.

  3. Only claim what we can show

    No numbers we have not measured, no capabilities we cannot demonstrate. Where something is unresolved, we say it is unresolved.

  4. Legible artifacts

    Outputs should be readable formats with published schemas, so other people can check our work without our software.

Contact

Reaching us

This site has no contact form, no newsletter, and no account system. Email reaches a person.

Privacy
privacy@yycore.ai
Security
security@yycore.ai
Notices
notices@yycore.ai