Update README.md

README.md

@@ -8,4 +8,27 @@ pinned: false
 license: mit
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# 📊 GEMS: Generalized Raman Classifier Webserver
+
+Welcome to the official interactive web application for **GEMS (Foundation Models for Universal, Data-efficient Classification of Raman Spectra)**.
+
+This Space provides a highly convenient, no-code graphical interface. It empowers researchers and practitioners to seamlessly perform fine-tuning, automated evaluation, and rapid predictions using the powerful GEMS foundation model.
+
+## 🔗 Quick Links
+
+* **Source Code:** Explore the core implementation, model architecture, and local deployment instructions in our [GitHub Repository](https://github.com/JunhanCai/GEMS).
+* **Datasets:** Access the complete, pre-processed multi-stage datasets (including pre-training, contrastive learning, and diverse downstream tasks) at our [Hugging Face Dataset Repo](https://huggingface.co/datasets/JunhanCai/GEMS-Raman-Dataset).
+
+## 🚀 How to Test it Out?
+
+You can test this web server immediately without preparing your own data! We have provided several pre-processed downstream datasets that are 100% compatible with this interface.
+
+**Step-by-step Guide:**
+1. Visit our [GEMS-Raman-Dataset](https://huggingface.co/datasets/JunhanCai/GEMS-Raman-Dataset) page.
+2. Download any downstream task subset that interests you (e.g., the `.npy` files from the `microplastic`, `skincancer`, or `Bacteria_ID` folders).
+3. Return to this Space and upload the downloaded `spectral.npy`, `labels.npy`, and `wavenumbers.npy` files into the **Fine-tune** panel.
+4. *(Optional)* Leave the `Pretrained Model` field empty! The server will automatically load our built-in foundation weights.
+5. Click **Start Fine-Tuning Job** to witness the automated AutoML pipeline and get your classification reports.
+
+---
+**Note:** Due to the hardware constraints of the free Hugging Face CPU tier, training may take some time. For production-level speed, we highly recommend pulling the Docker image or source code from our GitHub and deploying it locally on a GPU-enabled machine.