fix: use google/flan-t5-base - small, fast, 100% free model

config.py

@@ -37,8 +37,8 @@ CHROMA_COLLECTION_NAME = "document_embeddings"
 CHROMA_DISTANCE_METRIC = "cosine"
 
 # LLM Configuration - Hugging Face Inference API
-# Using a reliable model that works with HF Inference API free tier
-HF_MODEL = "mistralai/Mistral-7B-Instruct-v0.1"
+# Using a reliable FREE model that works with HF Inference API
+HF_MODEL = "google/flan-t5-base"  # Small, fast, 100% free, reliable
 HF_TOKEN = os.getenv("HF_TOKEN", "")
 
 # Retrieval Configuration