Upload README.md with huggingface_hub

README.md

@@ -19,14 +19,18 @@ tags:
 
 <div align="center">
 
+![TripAlchemy banner](assets/banner.png)
+
 # 🧪 TripAlchemy — AI Travel Experience Recommender
 
 ### *Set your travel vibe → get real matching experiences + one AI-woven itinerary*
 
-[![Gradio](https://img.shields.io/badge/Gradio-app-orange?logo=gradio)](https://gradio.app)
-[![Generation](https://img.shields.io/badge/generation-Qwen2.5--1.5B-2b7a78?logo=huggingface)](https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct)
-[![Embeddings](https://img.shields.io/badge/embeddings-MiniLM--L6-blue?logo=huggingface)](https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2)
-[![Dataset](https://img.shields.io/badge/dataset-10%2C396_experiences-ff6a3d?logo=huggingface)](https://huggingface.co/datasets/almador2002/tripalchemy-experiences)
+[![Gradio](https://img.shields.io/badge/Gradio-app-F97316?logo=gradio&logoColor=white)](https://gradio.app)
+[![Generation](https://img.shields.io/badge/Generation-Qwen2.5--1.5B-2b7a78?logo=huggingface&logoColor=white)](https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct)
+[![Embeddings](https://img.shields.io/badge/Embeddings-MiniLM--L6-4363d8?logo=huggingface&logoColor=white)](https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2)
+[![FAISS](https://img.shields.io/badge/Search-FAISS-8b3dff?logo=meta&logoColor=white)](https://github.com/facebookresearch/faiss)
+[![Dataset](https://img.shields.io/badge/Dataset-10%2C396_experiences-ff6a3d?logo=databricks&logoColor=white)](https://huggingface.co/datasets/almador2002/tripalchemy-experiences)
+[![Python](https://img.shields.io/badge/Python-3.10+-f7d54d?logo=python&logoColor=white)](https://www.python.org/)
 [![License](https://img.shields.io/badge/License-MIT-lightgrey)](https://opensource.org/licenses/MIT)
 
 **A hybrid recommender that turns six "vibe" sliders (or a free-text mood) into the best-matching real travel experiences across 30 cities — then weaves the top matches into a single AI-generated itinerary.**
@@ -124,17 +128,7 @@ final_score  =  α · text_embedding_similarity  +  (1 − α) · category_slide
 
 Even in pure-slider mode the sliders are converted into a synthetic sentence and embedded, so the embedding step is *never* bypassed. FAISS (`IndexFlatIP`) does exact cosine search over the whole collection.
 
-```
-User input (sliders or free text)
-        ↓
-   MiniLM encoder  ──►  512-D query vector
-        ↓
-   FAISS cosine search  vs. all 10,396 embeddings
-        ↓
-   Hybrid re-rank (embedding + slider dot-product) + hard filters (city, budget, energy)
-        ↓
-   Top-5 experiences  ──►  LLM itinerary narrator
-```
+![Pipeline](assets/pipeline.png)
 
 ### ✨ Part 4 — Generation