The Software Engineering Journey of Muhammad Zubair

In the rapidly evolving landscape of modern technology, the role of a Software Engineer has shifted from merely writing code to architecting complex, multi-layered ecosystems. My journey as an engineer, Muhammad Zubair, has been defined by a relentless pursuit of solving real-world problems through a combination of Generative AI, High-Performance Mobile Engineering, and Scalable Cloud Infrastructure.

This blog serves as a comprehensive technical retrospective of the major systems I have engineered—ranging from specialized Pashto AI models to enterprise-grade SaaS platforms and automated DevOps pipelines.

---

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 (Khatak) dialect of Pashto.

Khatta-ka-LLM: The world's first Khattak dialect AI, engineered by Muhammad Zubair.

The Technical Challenge

Standard Pashto LLMs do not recognize the unique linguistic markers of the Karak and Kohat regions. I curated a custom dataset of 2,000+ English-to-Khattak pairs and utilized the Qwen2 architecture with LoRA (Low-Rank Adaptation) via the Unsloth framework.

Key Engineering Feats:

• Phonetic "O" Shifts: Programmed the model to convert standard "A" sounds to the deep Khattak "O" (e.g., Asmaan ➡️ Asmon).
• The "Noon Ghunna" Rule: Implemented specific weights to handle nasal verb endings unique to the dialect.
• Performance: Reduced training loss from 3.44 to 0.22, creating a highly accurate regional translator.
• Live Demo: Try Khattak AI on Hugging Face

---

binAI Radar: AI-Powered Vision for the Blind

Navigating complex environments is a daily challenge for the visually impaired. I engineered binAI—a real-time, voice-controlled Android assistant that acts as a mobility radar.

The Hybrid Edge-to-Cloud Architecture

To ensure zero-latency safety, I designed a hybrid pipeline:

1. The Edge (Kotlin/CameraX): Processes frames locally using Google ML Kit to detect obstacles and calculate distances in feet.
2. The Brain (Vertex AI/Gemini 2.5): The phone injects mathematical sensor data into a prompt sent to a FastAPI backend.
3. The Fix: I solved a critical Android camera rotation bug using NumPy to fix C++ memory contiguity, ensuring face recognition works at any angle.

---

Bolo App: Engineering a WhatsApp-Scale Super-App

Bolo is an enterprise-grade Android platform that replicates the WhatsApp ecosystem. It features real-time chat, HD calling via Agora RTC, and encrypted status updates.

Engineering Highlights

I identified and solved a memory alignment issue for 64-bit architectures (the "16 KB Master Fix") by forcing native library extraction in the manifest.

The UI was built using Jetpack Compose and a custom Canvas API for the segmented status ring.

I also integrated BoloChef, a Generative AI module that processes ingredients to return recipes in English, Urdu, and Roman Urdu.

---

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 and Llama 3.1.

The "Clinical Board" Debate Arena

I architected a Directed Acyclic Graph (DAG) where four specialized agents (Diagnostic Lead, Clinical Reviewer, Medical Analyst, and Clinical Director) 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.

---

Enterprise SaaS: EduOps (School Management System)

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

The SaaS Core:

• Impersonation Mode: Engineered a secure way for Superadmins to assume tenant identities for debugging without compromising credentials.
• Financial Engine: A double-entry ledger system that tracks net profit, staff payroll, and automated fee defaulter alerts.
• Offline-First QR: Developed a mobile kiosk mode that scans student IDs offline and syncs to the cloud once connectivity is restored.
• Live Site: www.schoolsms.online

---

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.

High Availability with Terraform

I designed a "Black Friday-ready" architecture using Terraform that includes:

• Multi-AZ VPC: Spanning multiple availability zones for disaster recovery.
• ALB & ASG: An Application Load Balancer distributing traffic to an Auto Scaling Group of EC2 instances.
• State Locking: Implemented encrypted S3 Remote Backends with DynamoDB state locking to ensure team-ready, secure deployments.

DataOps CI/CD

To eliminate the "it works on my machine" problem, I built a Jenkins & Docker pipeline. It provisions ephemeral Python containers to validate and clean NLP data before it ever reaches a Machine Learning model, using strict exit codes to halt the pipeline if data anomalies are detected.

---

Creative Engineering: SimSea

To demonstrate a deep understanding of the fundamentals, I built SimSea, a 2D underwater shooter. Eschewing heavy engines like Unity, I used Vanilla JavaScript and HTML5 Canvas to build a custom 60FPS rendering engine, which I then packaged as a native Android APK using a Kotlin WebView wrapper.

---

Conclusion: The Engineering Philosophy

Whether it is fine-tuning an LLM for a rare dialect or architecting a multi-tenant SaaS, my philosophy remains the same: Build for scale, prioritize security, and always solve for the human experience.

As a Software Engineer, I am committed to pushing the boundaries of what is possible at the intersection of AI and Infrastructure.

---

About the Developer

I am Muhammad Zubair, a Software, DevOps, and Platform Engineer from Pakistan specializing in Agentic AI workflows, Cloud Infrastructure, and Full-Stack Systems.

• GitHub: Muhammad Zubair's Repositories (76+ Repos)
• Portfolio: mzubair.online
• SaaS Platform: EduOps (SMS)