docs: add Neo Ar dataset card

README.md

@@ -0,0 +1,55 @@
+---
+license: other
+license_name: tokenai-neo-ar-dataset-license
+license_link: https://huggingface.co/datasets/tokenaii/Neo-Ar-Dataset/blob/main/legal/DATASET_LICENSE.md
+language:
+- ar
+task_categories:
+- text-classification
+tags:
+- system-one
+- decision-model
+- rlcd
+- synthetic
+- tool-routing
+- arabic
+- arabic-only
+---
+
+# Neo Ar — Arabic Decision Model Dataset
+
+<p align="center"><img src="assets/neo-ar-cover.png" alt="Neo Ar — Arabic Decision Model Dataset" width="360"></p>
+
+This dataset contains synthetic Arabic-only decision records for training and
+evaluating Neo Ar. The model and dataset cards are written in English, but all
+state text, questions, labels, tool descriptions, and decision examples are in
+Arabic.
+
+## Initial target
+
+- 600,000 state records.
+- At least 1,800,000 individual decisions.
+- `choice`, `score`, and `noul` questions.
+- 40 Arabic-described tools and actions.
+- Train, calibration, and held-out test splits.
+
+The associated Neo Ar model contains 86,393,160 trainable parameters with the
+released Arabic tokenizer vocabulary.
+
+## Provenance
+
+Records are generated deterministically by the Arabic synthetic-data generator
+in the Neo Ar GitHub repository. No public dataset rows are copied into the
+training corpus. The final manifest will include counts, seed, split metadata,
+and SHA-256 hashes.
+
+## Links
+
+- Model: https://huggingface.co/tokenaii/Neo-Ar
+- Complete source and training repository: https://github.com/tokenaii/Neo-Ar
+
+Training logs and benchmark history are maintained on GitHub only and are not
+part of this dataset release.
+
+TokenAI is a non-profit startup founded in 2025 by Assem Sabry, based in
+Alexandria, Egypt. Contact: info@tokenai.llc — https://tokenai.llc.