Liodon AI is an open-source R&D collective exploring the intersection of deep learning, agentic systems, and NLP evaluation. Our repositories span:
- Agentic AI primitives (e.g.
torchagentic) — differentiable cognitive building blocks (memory, credit assignment, planning, exploration) designed to be importable by frameworks like LangChain, AutoGen, and CrewAI. - LLM evaluation tooling (e.g.
juryeval) — lightweight NLP/LLM evaluation with LLM-as-Judge, significance testing, prompt robustness analysis, and framework integrations (lm-eval-harness, DeepEval, Lighteval, LangChain). - Loss functions & optimization (e.g.
TorchCriterion) — modular, extensible deep learning loss functions in PyTorch. - On-device & efficient ML (e.g.
nanonlp) — hardware-aware optimization for HuggingFace Transformers. - Speech & embedding evaluation (e.g.
semanticwer,torchembed) — evaluation metrics for speech recognition and embedding quality. - Prompt engineering tooling (e.g.
promptgrad) — prompt sensitivity analysis and gradient-based debugging.
Liodon AI is not a company. It is not incorporated, registered, or organized as any form of legal entity (corporation, LLC, partnership, or otherwise) in any jurisdiction. There is no revenue, no fundraising, no equity, no compensation, and no commercial operations associated with this organization.
This GitHub organization exists solely as a collective open-source project — a home for experimental AI research and tooling developed in the open. All repositories under this organization are published under permissive open-source licenses (MIT / Apache 2.0) and are available for anyone to use, modify, and distribute free of charge.
- Not a business. No goods or services are sold. No revenue is generated or accepted.
- Not an employer. No employment or contractor relationships exist.
- Not an investment vehicle. No equity, tokens, or financial instruments have been or will be offered.
- Not a legal entity. This organization has no legal personality, no registered address, and no governing board.
Contributions are welcome under each project's license terms. By contributing, you agree that your contributions are made voluntarily without expectation of compensation or any legal relationship with this organization.
All software, documentation, and other content in this organization is provided "as is," without warranty of any kind, express or implied. The individuals maintaining these projects assume no liability for any claims, damages, or other liabilities arising from the use of the software.