Muhammad Zubair 👋

Khattak AI: Preserving Dialects with a Pashto LLM

Standard AI models often fail to capture the rich, localized grammar of rural dialects. To solve this, I engineered Khatta-ka-LLM, the world's first AI language model fine-tuned specifically for the Khattak dialect of Pashto.

• ▹ The Tech: Qwen2 architecture with LoRA via the Unsloth framework.
• ▹ Phonetic Shifts: Programmed the model to convert standard "A" sounds to the deep Khattak "O" (e.g., Asmaan ➡️ Asmon).
• ▹ Performance: Reduced training loss from 3.44 to 0.22.

Enterprise SaaS: EduOps

Building for one school is easy; building for thousands requires a robust Multi-Tenant SaaS Architecture. EduOps is a high-performance ERP built with TypeScript that automates everything from fee collection to QR-based attendance.

• ▹ Impersonation Mode: Engineered a secure way for Superadmins to assume tenant identities for debugging.
• ▹ Financial Engine: A double-entry ledger system tracking net profit and payroll.
• ▹ Offline-First QR: Mobile kiosk mode that scans student IDs offline and syncs to the cloud.

Multi-Agent Systems: HealthConnect Pro

In healthcare, a single AI prompt is dangerous. For HealthConnect Pro, I moved beyond linear LLM calls to a Multi-Agent System (MAS) using LangGraph, FastMCP, and Llama 3.3 70B.

I architected a Directed Acyclic Graph (DAG) where four specialized agents debate a patient's symptoms. This "self-correction" mechanism dropped hallucinations to near zero. I also implemented a safe_invoke fallback that switches from Llama 70B to 8B during API rate-limiting to ensure 100% uptime. Deployed on Red Hat OpenShift using Tekton CI/CD.

DevOps & Infrastructure as Code (IaC)

A great application is only as good as the infrastructure it runs on. I specialize in architecting resilient, automated environments on AWS and Kubernetes.

I designed a "Black Friday-ready" architecture using Terraform featuring a Multi-AZ VPC, Application Load Balancers, and Auto Scaling Groups. To eliminate the "it works on my machine" problem, I built a Jenkins & Docker pipeline that provisions ephemeral Python containers to validate NLP data before it reaches ML models.

binAI Radar, Bolo App & Game Dev

Beyond cloud and AI, I engineer high-performance mobile and web solutions. binAI is a real-time, voice-controlled Android assistant acting as a mobility radar for the visually impaired, utilizing a hybrid Edge-to-Cloud architecture.

Bolo is an enterprise-grade Android platform replicating the WhatsApp ecosystem, featuring real-time chat, HD calling, and a Generative AI module. I also build interactive web experiences like SimSea, a 2D game built entirely with Vanilla JavaScript and HTML5 Canvas.