add tools support

changing response way llm

changing response way

Update system model to Gemma 3 1B Instruct and humanize responses

feat: implement semantic chunking, dual-retrieval cross-lingual search, and direct Q&A bypass routing

Fix Hindi/Hinglish language output alignment using retrieval translation pipeline

fix: add threading lock synchronization to prevent concurrent model inference crashes

feat: optimize response time by offloading translation to deep-translator and reducing ctx/retrieval k size

feat: implement multilingual query translation and language-specific prompting

feat: implement multilingual RAG pipeline for English, Hindi, and Hinglish

Optimize Hugging Face Space: add eager model loading, reduce max tokens, fix stop tokens, limit CPU threads

feat: implement complete RAG-based AI engine with Ollama fallback and vector search support

Optimize startup speed and query latency

Initial clean release