🛡️ QueryShield — Multilingual Prompt Optimizer

Fine-tuned Qwen2.5-1.5B that rewrites raw user queries into expert-level instruction prompts for downstream LLMs.

Supports 5 languages: English · Uzbek · Russian · Kazakh · Karakalpak
Supports cross-lingual routing: write in one language, get instructions for another.

⚠️ Running on CPU — generation takes ~30–60 seconds. Please be patient.

📦 Dataset · 🤖 Model · 📓 Kaggle Demo

Raw User Question

Input Language

Output Language

Expert Role

Max output tokens

100 600

Optimized Prompt (instruction for downstream LLM)

📌 Example queries (click to load)

Raw User Question	Input Language	Output Language	Expert Role

How it works:
QueryShield sits between the user and the main LLM. It takes a raw query and outputs a structured instruction prompt — including role, tone, format, edge cases, and language routing instructions.

Built with ❤️ by nickoo004