| --- |
| title: World Cup Storyteller |
| emoji: ⚽ |
| colorFrom: green |
| colorTo: yellow |
| sdk: gradio |
| sdk_version: 6.12.0 |
| app_file: app.py |
| pinned: false |
| license: mit |
| --- |
| |
| # ⚽ World Cup Storyteller |
|
|
| An interactive NLP app that generates vivid, narrative-driven stories about FIFA World Cup editions and team journeys — built for NLP Homework 4 (ARI 525). |
|
|
| ## What it does |
|
|
| Pick a World Cup year, a team, write a free-form prompt (or leave it blank), and choose one of three NLP approaches to generate a story: |
|
|
| - **Zero-Shot** — Raw match data + instruction, no examples |
| - **Few-Shot** — Guided by hand-crafted story examples baked into the prompt |
| - **RAG** — Retrieves the most relevant context from an embedded knowledge base before generating |
|
|
| ## Data |
|
|
| FIFA World Cup dataset (1930–2014) from Kaggle — covers all matches, scores, teams, and edition summaries. |
|
|
| ## Stack |
|
|
| - 🦙 **Llama 3 8B** via Groq API (LLM backbone) |
| - 🔍 **all-MiniLM-L6-v2** (sentence-transformers) for RAG embeddings |
| - 📦 **FAISS** for vector similarity search |
| - 🎛️ **Gradio** for the UI |
