fix: use microsoft/Phi-3-mini-4k-instruct - officially supported free model

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

fix: switch to Mistral-7B model for better HF Inference API compatibility

fix: use google/flan-t5-large model for HF Inference API

feat: migrate from Ollama to Hugging Face Inference API

Initial commit: RAG system with local LLM