Rahnuma (راہنماء) — Friendly Urdu AI Assistant

Rahnuma is a production-quality Urdu AI assistant designed to help users with general conversations, text rewriting, and content summarization. It is built with a focus on Responsible AI, incorporating multi-stage safety guardrails to ensure it does not provide medical, legal, or religious advice.

Screen 2	Screen 3

• Native Urdu Fluency: Optimized using the Qalb-1.0-8B-Instruct model for natural Pakistani Urdu.
• Advanced Orchestration: Conversation logic managed by LangGraph for robust state management.
• Safety First: Integrated keyword filtering and LLM-based intent classification to block sensitive domains.
• Premium UI: Modern, responsive Right-to-Left (RTL) chat interface built with accessibility in mind.

• LLM Core: Qalb-1.0-8B-Instruct (HuggingFace Transformers)
• Quantization: BitsAndBytes (4-bit NF4 quantization for efficient GPU memory usage)
• Agent Framework: LangGraph (Stateful orchestration)
• Backend API: FastAPI (High-performance asynchronous API)
• Inference Engine: PyTorch & Accelerate
• Frontend: Custom HTML5, CSS3 (with Google Fonts Integration), and Vanilla JavaScript
• Web Server: Uvicorn

The assistant is strictly constrained to avoid high-stakes advice. The pipeline includes:

1. Classify Intent: Detects if the user is asking for chat, rewriting, or blocked topics.
2. Safety Check: Immediate refusal of medical, legal, or religious queries.
3. Post-Processing: Ensures output quality and removes repetitive patterns.

rahnuma/
 ├─ app/
 │   ├─ main.py      # FastAPI Server & Routes
 │   ├─ graph.py     # LangGraph Pipeline logic
 │   ├─ model.py     # Model loading & Inference
 │   ├─ safety.py    # Heuristic guardrails
 │   └─ prompts.py   # All Urdu system prompts
 ├─ ui/
 │   └─ index.html   # Front-end chat UI
 ├─ requirements.txt # Project dependencies
 └─ README.md        # Documentation

Ensure the model files are located at ./models/Qalb-1.0-8B-Instruct.

pip install -r requirements.txt

python -m rahnuma.app.main

Access the interface at http://localhost:8001.

AI Engineer | RAG Systems | LLM Chatbots | AI Agents

I build production-grade AI systems that help businesses:

• RAG Systems: Chat with private data securely.
• AI Agents: Automate complex multi-step workflows.
• Custom LLM Solutions: Fine-tuned and optimized for specific niche tasks.
• Python-based AI Backends: Scalable, reliable, and high-performance.

• 📧 Email: [email protected]
• 🔗 LinkedIn: Hassan Khan
• 🔗 Upwork: Work with me on Upwork

Disclaimer: Rahnuma is an AI assistant and is not a substitute for professional human consultation.