👨‍💻 Name:        Mussa Khan
🎯 Role:        AI Engineer  •  AI & ML Instructor  •  Open-Source Developer
🏫 Education:   BS Artificial Intelligence  (5th Semester  •  GPA 3.2)
🌍 Based in:    Islamabad, Pakistan  🇵🇰
🏢 Currently:   AI & ML Instructor @ HectaSolution
🧠 Focus:       Agentic AI  •  LLM Applications  •  Computer Vision  •  AI Automation
💡 Mission:     "Translate cutting-edge AI research into shippable, real-world products
                 — and teach the next generation to do the same."
📫 Reach:       raja.musa.khan035@gmail.com
⚡ Fun fact:    I built a portable AI agent that runs entirely from a USB drive.

I'm an AI Engineer specializing in Agentic AI systems, LLM-powered applications, and production-grade Computer Vision. By day I design and ship AI products — chatbots, RAG pipelines, vision systems, and automation workflows. By night I teach those same skills to the next generation of AI builders as an instructor at HectaSolution.

My work sits at the intersection of research and deployment: I love taking a paper or a model and turning it into something people can actually use — whether that's a real-time stress detector running at 93% AUC, an interview copilot powered by emotion-aware LLMs, or a self-contained AI agent that fits in your pocket on a USB stick.

Languages

AI / Machine Learning

Computer Vision

Frameworks & Tools

DevOps & Platforms

Every project below ships production-ready code, documentation, and reproducible results. Click any title to dive in.

A fully portable, 100% offline AI agent that lives on a USB drive.

A three-tier agentic runtime designed for ultimate portability and privacy — plug the USB into any Windows / macOS / Linux machine, and your personalized LLM agent boots up with zero installation and zero telemetry.

🏗️ Architecture

• The Engine → Ollama (localhost:11434) running quantized models like qwen2.5:0.5b / llama3
• The Brain → Python HTTP Agent API (localhost:11500) handling system prompt, memory, and tool execution
• The Frontend → Static HTML / JS dashboard (localhost:8888) for chat UI

✨ Highlights   Portable Agentic AI   Local LLM   Tool Calling   Sandbox Security   Cross-Platform

The world's first emotion-aware interview simulator that adapts to your stress level in real time.

HireSense AI combines RAG personalization, client-side emotion detection, and adaptive LLM behavior to create an interview experience that feels uncannily real — getting tougher when you're confident and more supportive when you're stressed.

🔥 Standout Features

• 📄 RAG personalization from your resume + target job description
• 😐➡️😨 Real-time facial emotion detection via TensorFlow.js + face-api.js (client-side, private)
• 🎙️ Voice-driven interaction (Web Speech Recognition + TTS)
• 📊 AI report card with skill-gap analysis, non-verbal cues (eye contact, posture)
• 💻 Integrated coding + whiteboard + Live Interview Copilot
• 🏢 Company-specific prep for Google, Amazon, and more

A real-time facial expression sensor that knows if you're nailing the interview — or panicking inside.

Instead of using noisy 7-class emotion models, I re-engineered the FER2013 dataset into a binary contextual problem (Confident vs Stressed) and trained a custom CNN with Spatial Attention Modules that focuses on the most informative facial regions (eyes & mouth).

📈 Results

🧪 Engineering

• Custom CNN + Spatial Attention + BatchNorm + Dropout
• Mixed-Precision Training on T4 GPU (Google Colab)
• Exported to Keras / TF SavedModel / TFLite for flexible deployment
• 90+ epochs of training on 29,262 re-mapped images

A production-grade, cross-platform ALPR system written in modern C++ for traffic, parking, and security deployments.

Built for real-time inference on edge hardware, this system uses YOLOv8 exported to ONNX and served through OpenCV's DNN module in C++ — squeezing out every millisecond of latency for live traffic feeds and IP cameras.

⚡ Highlights

• 🎥 Multi-source input: images, videos, webcam, IP cameras
• 🪟🐧🍎 Cross-platform — Windows, Linux, macOS
• 🚀 ONNX-optimized inference pipeline
• 🏭 Production-ready modular C++14 codebase + CMake build

An all-in-one facial analysis toolkit with both GUI and CLI workflows.

A comprehensive computer-vision Swiss-Army-knife built on OpenCV + Tkinter, covering everything from age/gender prediction to head-pose estimation, facial symmetry scoring, glasses detection, privacy blurring, and artistic filters.

🧩 Modules

• Core Detection → Face, Age, Gender, Emotion, Smile intensity
• Advanced Analysis → Head Pose, Face Shape, Glasses Detection, Symmetry %
• Processing → Sepia / B&W / Cartoon filters, Face Tracking, Privacy Blur
• Utilities → Batch processing, dimensional measurements, GUI + CLI

Open for freelance, consulting, and collaboration — particularly with startups, AI-first companies, and educational platforms.

🐍 Contribution Snake

+ 🤖  Production-grade Agentic AI systems (tool-calling, planning, memory)
+ 🧠  Local-first LLM deployments with Ollama + on-device quantization
+ 🎥  Real-time multimodal Computer Vision (vision + voice + emotion)
+ 🔗  Retrieval-Augmented Generation (RAG) for domain-specific assistants
+ 📚  Open-source educational AI tooling for students worldwide
! 🚧  Long-term: Building an AI-powered learning platform for Pakistani students

I'm always open to freelance AI projects, research collaborations, teaching opportunities, and open-source contributions. If you're building something exciting in Agentic AI, LLMs, or Computer Vision, I'd love to hear from you.

💭 Dev Quote of the Day

_{⭐ If you like my work, consider starring some repositories — it genuinely makes my day.}