docs: update README with configurable embedding models

- Add embedding model comparison table in README
- Update features list to mention configurable embedding models
- Add new section 'Configurable Embedding Models' with usage examples
- Document available models: MiniLM, MPNet, and Gemma

README.md

@@ -8,7 +8,7 @@ An autonomous AI player for MUD (Multi-User Dungeon) text-based games.
 - 🗺️ **Auto-Mapping**: Builds a complete map of the MUD world with hybrid storage (SQLite + Memvid)
 - ⚔️ **Combat AI**: Intelligent combat decisions, target selection, and survival tactics
 - 📜 **Quest System**: Automatically detects, plans, and completes quests
-- 🔍 **Semantic Search**: Find locations using natural language queries
+- 🔍 **Semantic Search**: Find locations using natural language queries with configurable embedding models (MiniLM, MPNet, Gemma)
 - 📊 **Progression Optimizer**: Finds optimal grinding spots and equipment upgrades
 - 🧠 **LLM Integration**: Context-aware responses to players using local LLMs (Ollama)
 - 🌐 **Web Dashboard**: Real-time monitoring dashboard with WebSocket updates
@@ -129,6 +129,26 @@ monitoring:
   port: 8080
 ```
 
+## Configurable Embedding Models
+
+Choose the best embedding model for semantic room search:
+
+```yaml
+mapper:
+  semantic_search:
+    # Available models:
+    # - "sentence-transformers/all-MiniLM-L6-v2" (fast, 22MB) - default
+    # - "sentence-transformers/all-mpnet-base-v2" (better quality, 418MB)
+    # - "google/embeddinggemma-300m" (state-of-the-art, ~300MB)
+    model: "google/embeddinggemma-300m"
+```
+
+| Model | Size | Speed | Quality | Use Case |
+|-------|------|-------|---------|----------|
+| **all-MiniLM-L6-v2** | 22MB | ⭐⭐⭐⭐⭐ | ⭐⭐⭐ | Low resource, fast queries |
+| **all-mpnet-base-v2** | 418MB | ⭐⭐⭐ | ⭐⭐⭐⭐ | Better accuracy, balanced |
+| **google/embeddinggemma-300m** | ~300MB | ⭐⭐⭐⭐ | ⭐⭐⭐⭐⭐ | Best quality, requires transformers>=4.40 |
+
 Access the dashboard at `http://127.0.0.1:8080`
 
 **Features:**