# Pak-LLM — Pakistan's Sovereign AI Platform
> Real-time Urdu Voice AI, custom Nastaliq BPE tokenizer, on-premise enterprise LLM deployments, and Islamic ethical governance.

## What is Pak-LLM?

Pak-LLM is Pakistan's sovereign AI ecosystem built and operated by SBF Consultancy (Karachi). It is NOT a wrapper around ChatGPT or any foreign cloud AI — it is a ground-up sovereign platform with its own tokenizer, on-premise inference infrastructure, and ethical governance layer.

Key differentiators:
- **Custom Nastaliq BPE Tokenizer**: 95% subword compression on Urdu, Sindhi, Pashto, Punjabi Shahmukhi, Balochi, and Saraiki compared to GPT-4o/LLaMA standard tokenizers.
- **Real-Time Urdu Voice AI**: Bidirectional spoken Urdu conversations with sub-300ms speech-to-speech latency using Faster-Whisper ASR + Coqui XTTS-v2.
- **On-Premise Enterprise Deployments**: Air-gapped LLM deployments in Karachi data centers for banking, defense, government, and telecom.
- **NIST FIPS 203 Compliance**: Post-quantum cryptography (ML-KEM-768) for data at rest and in transit.
- **Islamic Ethical AI (Amanah Gate)**: Shariah-compliant content interception using Holy Quran (114 Surahs) RAG, Kutub al-Sittah Hadith dataset, and Maqasid al-Shariah alignment.

## Subscription Plans (PKR)

1. Free Awaam — PKR 0/mo (15 min voice, 500K tokens)
2. Talib-e-Ilm (Student) — PKR 990/mo (60 min voice, 2M tokens)
3. Freelancer Pro — PKR 2,490/mo (250 min voice, 5M tokens)
4. Kasib / SME — PKR 4,990/mo (500 min voice, 10M tokens)
5. Tajir / E-commerce — PKR 7,990/mo (800 min voice, 20M tokens)
6. Alim / Shariah — PKR 9,990/mo (1,200 min voice, 30M tokens)
7. Wakeel / Legal — PKR 12,490/mo (1,500 min voice, 40M tokens)
8. Idara / Government — PKR 24,990/mo (3,000 min voice, 75M tokens)
9. Khalij / Enterprise — PKR 44,990/mo (3,500 min voice, 150M tokens + on-premise option)

## Core Services

- Urdu Voice AI: https://pak-llm.com/how-to-use
- AI Packages & Pricing: https://pak-llm.com/packages
- Language Support (Urdu, Sindhi, Pashto, Punjabi, Balochi, Saraiki, Arabic): https://pak-llm.com/languages
- On-Premise Architecture: https://pak-llm.com/architecture
- Data Center (Karachi sovereign node): https://pak-llm.com/datacenter
- Islamic Ethical Framework: https://pak-llm.com/islamic-framework
- Developer API Docs: https://pak-llm.com/docs/api
- Blog: https://pak-llm.com/blogs
- FAQ: https://pak-llm.com/faq
- Contact: https://pak-llm.com/contact

## Technical Stack

- LLM Inference: vLLM + LLaMA 3 (fine-tuned on Urdu/Perso-Arabic corpus)
- ASR: Faster-Whisper (Pakistani Urdu fine-tuned)
- TTS: Coqui XTTS-v2 (Urdu neural synthesis)
- Tokenizer: Custom Nastaliq BPE (PyTorch / SentencePiece)
- Vector DB: PostgreSQL pgvector (Qdrant optional)
- RAG: Crawl4AI + hybrid BM25+vector retrieval
- Security: NIST FIPS 203 ML-KEM-768, NIST SP 800-88, Pakistan PDPB Draft
- Platform: Next.js 15, TypeScript, Firebase Auth, PostgreSQL

## Contact

- Website: https://pak-llm.com
- Enterprise Sales: info@sbf-consultancy.net
- Phone: +92-322-247-6072
- Address: B-48, Double Road, Phase 2, Gulshan-e-Hadeed, Malir, Karachi, Sindh 75010, Pakistan
- GitHub: https://github.com/saado1o/pak-llm

## Crawler Policy

This file is maintained for AI language model crawlers (GPTBot, PerplexityBot, ClaudeBot, Google-Extended, OAI-SearchBot, Gemini-Google, Bytespider, cohere-ai).
All public marketing pages are permitted. Authenticated user areas (/chat, /admin, /api) are excluded.
Full robots.txt: https://pak-llm.com/robots.txt
Sitemap: https://pak-llm.com/sitemap.xml