Org card redesign in the Netra Lab website style

README.md

@@ -7,32 +7,56 @@ sdk: static
 pinned: false
 ---
 
+<img src="https://huggingface.co/spaces/Netra-Lab/README/resolve/main/assets/hero.png" width="100%" alt="Netra Lab. Open AI for the Khmer language, built in the open. Netra Lab is an independent, non-commercial research lab. We build small open models, datasets and tools for Khmer, and publish the code, weights and a full write-up for each one, free." />
+
 <p align="center">
-  <img src="https://huggingface.co/spaces/Netra-Lab/README/resolve/main/assets/header.png" width="260" alt="Netra Lab" />
+  <a href="https://netra-lab.netlify.app/"><b>Website</b></a> ·
+  <a href="https://netra-lab.netlify.app/research"><b>Research</b></a> ·
+  <a href="https://netra-lab.netlify.app/playground"><b>Playground</b></a> ·
+  <a href="https://github.com/netra-ai-lab"><b>GitHub</b></a>
 </p>
 
-**Make Khmer a first-class language for AI.** Netra Lab is an independent, non-commercial research lab. We build
-small open models, datasets and tools for Khmer, sized for the devices people already own, and publish the code,
-weights and data so no one working on Khmer has to start from zero.
+## Latest research
+
+<p>
+  <a href="https://netra-lab.netlify.app/research/sota"><img src="https://huggingface.co/spaces/Netra-Lab/README/resolve/main/assets/research-sota.png" width="49%" alt="Sota: Khmer and English speech recognition, pruned to a fifth of its size. 73M / 204M parameters, 1.77% CER, Apache-2.0." /></a>
+  <a href="https://netra-lab.netlify.app/research/netra-ocr"><img src="https://huggingface.co/spaces/Netra-Lab/README/resolve/main/assets/research-netra-ocr.png" width="49%" alt="Netra OCR 1: a 17M-parameter model for reading printed Khmer and English text. 1.3M training images, MIT." /></a>
+</p>
+<p>
+  <a href="https://netra-lab.netlify.app/research/netra-mt"><img src="https://huggingface.co/spaces/Netra-Lab/README/resolve/main/assets/research-netra-mt.png" width="49%" alt="Netra MT 1: a compact bidirectional encoder-decoder transformer for English–Khmer translation. 77M parameters, 220M training tokens, MIT." /></a>
+  <a href="https://netra-lab.netlify.app/research/akara-alpha"><img src="https://huggingface.co/spaces/Netra-Lab/README/resolve/main/assets/research-akara-alpha.png" width="49%" alt="AkaraAlpha: a small model that turns romanized Khmer into Khmer script. 78.7K parameters, 15.07% CER, MIT." /></a>
+</p>
 
-Every model here ships as ONNX with a pip package that runs it on a CPU, without PyTorch.
+## Open releases
 
-**Models**
+Small open models for Khmer, sized to run on an ordinary laptop, with the weights free to download and fine-tune.
 
-| Model | What it does | Use it |
+| Model | Task | Get it |
 | --- | --- | --- |
-| [Netra OCR](https://huggingface.co/Netra-Lab/Netra-OCR) | Khmer and English OCR: text-line detection and a 19M-parameter line recognizer | [`pip install netra-ocr`](https://pypi.org/project/netra-ocr/) · [GitHub](https://github.com/netra-ai-lab/Netra-OCR) |
-| [Netra NMT](https://huggingface.co/Netra-Lab/Netra-NMT) | English↔Khmer translation, 77M parameters, both directions in one model | [`pip install netra-nmt`](https://pypi.org/project/netra-nmt/) · [GitHub](https://github.com/netra-ai-lab/Netra-NMT) · [Demo](https://huggingface.co/spaces/Darayut/Netra-NMT) |
+| [**Netra OCR**](https://huggingface.co/Netra-Lab/Netra-OCR) | Printed Khmer and English text: line detection and recognition | [`pip install netra-ocr`](https://pypi.org/project/netra-ocr/) · [GitHub](https://github.com/netra-ai-lab/Netra-OCR) |
+| [**Netra NMT**](https://huggingface.co/Netra-Lab/Netra-NMT) | English↔Khmer translation, both directions in one 77M model | [`pip install netra-nmt`](https://pypi.org/project/netra-nmt/) · [GitHub](https://github.com/netra-ai-lab/Netra-NMT) · [Demo](https://huggingface.co/spaces/Darayut/Netra-NMT) |
+| **Sota** | Khmer and English speech recognition | [sota-large](https://huggingface.co/Darayut/sota-large) · [sota-base](https://huggingface.co/Darayut/sota-base) · [GitHub](https://github.com/netra-ai-lab/SOTA) |
+| **AkaraAlpha** | Romanized Khmer to Khmer script | [GitHub](https://github.com/netra-ai-lab/Romanized-Khmer-to-Khmer-Transliteration) |
 
-**Datasets**
+The Khmer data we build to train and test them, published so the next project doesn't have to start from zero.
 
 | Dataset | Contents |
 | --- | --- |
-| [bilingual-en-km](https://huggingface.co/datasets/Netra-Lab/bilingual-en-km) | 4.26M English↔Khmer translation examples, both directions: Netra NMT's training corpus |
-| [khmer-english-pairs-raw](https://huggingface.co/datasets/Netra-Lab/khmer-english-pairs-raw) | 200K synthetic Khmer-English sentence pairs (legal, literary, governmental), MIT |
+| [bilingual-en-km](https://huggingface.co/datasets/Netra-Lab/bilingual-en-km) | 4.26M English↔Khmer translation examples: Netra NMT's training corpus |
+| [khmer-english-pairs-raw](https://huggingface.co/datasets/Netra-Lab/khmer-english-pairs-raw) | 200K synthetic Khmer-English sentence pairs (legal, literary, governmental) |
 | [khob-eval](https://huggingface.co/datasets/Netra-Lab/khob-eval) | 325 Khmer text lines from scanned government documents: an OCR test set |
 | [khmer-printed-words](https://huggingface.co/datasets/Netra-Lab/khmer-printed-words) | 5,000 images of short Khmer words in 10 fonts: an OCR test set for short input |
 | [khmer-document-synthetic-low-res](https://huggingface.co/datasets/Netra-Lab/khmer-document-synthetic-low-res) | 100,000 synthetic Khmer document text-line images with their text |
 | [khmer-scene-text-synthetic-contrast](https://huggingface.co/datasets/Netra-Lab/khmer-scene-text-synthetic-contrast) | 102,500 synthetic Khmer scene-text images with their text |
 
-Code, training recipes and results: [github.com/netra-ai-lab](https://github.com/netra-ai-lab) · Website: [netra-lab.netlify.app](https://netra-lab.netlify.app/)
+<img src="https://huggingface.co/spaces/Netra-Lab/README/resolve/main/assets/band.png" width="100%" alt="More than 16 million people speak Khmer. Most AI still treats it as an afterthought. There's little public data to train on, and tools built for English stumble on a script where consonants stack and words run together without spaces. Make Khmer a first-class language for AI: we build the models and test sets Khmer speakers need, sized for the devices they already own, and publish all of it so no one working on Khmer has to start from zero." />
+
+## How we work
+
+- **Free and open.** Every model, dataset and line of code is open source, under licenses that allow commercial use. No account, API key or credit card needed.
+- **Khmer first.** Every model is designed around how Khmer is written and spoken: consonants stack, vowels attach around them, and words run together without spaces.
+- **Honest about what works.** Each release ships with a written report and its test data, failures included, so you can check the numbers yourself.
+- **Runs on edge devices.** Models are sized for an ordinary laptop or phone.
+- **Built to be built on.** The goal is a healthier Khmer AI ecosystem, not a moat. If someone trains something better on top of this work, that's the point.
+
+<p align="center"><sub>Netra Lab is independent and non-commercial · <a href="https://netra-lab.netlify.app/about">About the lab</a> · <a href="https://netra-lab.netlify.app/contact">Get involved</a></sub></p>