Rewrite org card in LightOn-style platform pitch

README.md

@@ -7,33 +7,15 @@ sdk: static
 pinned: false
 ---
 
-**Context212** builds document intelligence for teams that need accurate OCR and RAG over real institutional paperwork — including Arabic-first forms, invoices, and handwriting-heavy pages.
+<b>Document AI for the Enterprise</b>
 
-### Featured: Alhazen-OCR
+Context212 is the platform for document intelligence teams:
+🚀 easy to deploy — parse, extract, search, and ask with one API
+🎨 built for real institutional paperwork, including Arabic-first OCR
+🔒 on-premise or sovereign cloud, so your data and IP stay yours
 
-Compact QLoRA adapter on Qwen3-VL-2B for Arabic institutional OCR. Held-out character error drops from **0.79 → 0.33 CER** vs the untuned base (greedy, n=50).
+Learn more at <a href="https://context212.com">context212.com</a> 💡 Get in touch if you want help shipping document AI for regulated teams.
 
-| | |
-|---|---|
-| Model | [`context212/alhazen-ocr`](https://huggingface.co/context212/alhazen-ocr) |
-| Dataset | [`context212/context212-alhazen-ocr`](https://huggingface.co/datasets/context212/context212-alhazen-ocr) |
-| Code | [`github.com/context212/atlas-ocr`](https://github.com/context212/atlas-ocr) |
-| Write-up | [Alhazen-OCR release post](https://context212.com/blog/alhazen-ocr) |
+<b>About Context212</b>
 
-```bash
-vllm serve Qwen/Qwen3-VL-2B-Instruct \
-  --enable-lora \
-  --lora-modules alhazen=context212/alhazen-ocr \
-  --limit-mm-per-prompt '{"image": 1}'
-```
-
-### On the Hub
-
-- **Models** — [browse `context212`](https://huggingface.co/context212/models)
-- **Datasets** — [browse datasets](https://huggingface.co/context212/datasets)
-
-### Links
-
-- Website: [context212.com](https://context212.com)
-- Docs: [developers.context212.com](https://developers.context212.com)
-- GitHub: [github.com/context212](https://github.com/context212)
+We build document AI primitives for enterprises and institutions that need control over their data. On the Hub we release open models and datasets for institutional OCR — starting with <a href="https://huggingface.co/context212/alhazen-ocr">Alhazen-OCR</a>, a compact QLoRA adapter for Arabic forms, invoices, and handwriting-heavy pages, trained on license-clean data (<a href="https://huggingface.co/datasets/context212/context212-alhazen-ocr">context212-alhazen-ocr</a>).