Upload backend_content/app/llm.py with huggingface_hub

Upload backend_content/app/bots/telegram.py with huggingface_hub

Upload backend_content/app/bots/telegram.py with huggingface_hub

Upload backend_content/app/config.py with huggingface_hub

Upload backend_content/app/bots/telegram.py with huggingface_hub

Upload backend_content/app/medical_knowledge.py with huggingface_hub

Upload backend_content/app/bots/telegram.py with huggingface_hub

Upload backend_content/app/bots/telegram.py with huggingface_hub

Upload backend_content/app/config.py with huggingface_hub

Upload backend_content/app/config.py with huggingface_hub

Upload backend_content/app/llm.py with huggingface_hub

Upload backend_content/app/bots/telegram.py with huggingface_hub

fix: simplify inline response (remove Markdown and reply_to)

fix: return Telegram inline-response from webhook endpoint

fix: use Telegram inline-response (no outbound connection needed)

debug: expose full error in webhook handler

fix: rewrite telegram bot using direct httpx (no Application.initialize)

fix: force Arabic responses in system prompt

fix: use Qwen2.5-7B-Instruct model

fix: use router.huggingface.co/v1 endpoint

fix: update HF inference URL to router.huggingface.co

fix: use HuggingFace inference endpoint in llm.py

fix: switch AI to HuggingFace Inference API (free)

fix: non-blocking Telegram setup with lazy init and webhook retries

fix: add missing settings import in telegram.py

fix: copy requirements from backend_content

feat: switch AI provider to ZhipuAI (z.ai GLM-4-Flash)

feat: deploy FastAPI backend to HF Docker Space

initial commit