AAME
AAME is an AI research and systems company. We build small, inspectable models for decisions and the systems around them, and we publish each release with its interfaces, evaluation protocol, measured results and provenance.
What we research
Efficient decision models
Small, specialised models that choose among typed answers, trained from scratch and fast on a CPU.
Probabilistic AI systems
Outputs as probability distributions whose quality is measured and published, so a system can decide when to act and when to defer.
Evaluation and reproducibility
Sealed evaluation rules, evidence classes for every number, and releases anyone can verify byte for byte.
How we work
Sealed evaluation
The protocol is fixed before the test split is opened. The sealed release evaluation records the original result; later verification and diagnostics are identified separately.
Evidence classes
Every number we publish is marked measured by us, reproduced by us, or reported externally. Comparisons identify our measurements, reproduced checkpoints and figures reported by their publishers.
Verifiable releases
Checksums for every file, provenance for every model file, a public runtime in plain Python, and a behavioral-equivalence record between the public package and our internal frozen model.
Founder
AAME was founded by Sai Tilak Pally. He leads the Primus programme and is the author of record for each Primus release.
Primus
Primus is our research and model programme. Its first public release is Primus Decision 0.1. Our longer-term goal connects memory, recurrent graph reasoning, learning, grounding, planning and tool interfaces. We describe these research directions publicly while protecting unreleased implementation details.
Contact
For research, collaboration or press:
connect@theaame.com