class MuhammadFezan:
def __init__(self):
self.username = "H0NEYP0T-466"
self.name = "Muhammad Fezan"
self.location = "Lahore, Pakistan 🇵🇰"
self.role = "AI Systems Engineer & Full-Stack Developer"
self.university = "Lahore Garrison University"
self.degree = "BSCS — 6th Semester"
self.email = "fezan1029@gmail.com"
@property
def tech_stack(self):
return {
"ai_ml": ["Multi-Agent Systems", "RAG Pipelines", "LLM Fine-tuning (LoRA/QLoRA)",
"Computer Vision", "Deep Learning", "Self-Supervised Learning (DINOv2)",
"Medical Segmentation (U-Net, SAM-2)", "FAISS", "Agentic Workflows"],
"llm_providers": ["OpenAI", "Anthropic", "Mistral", "Groq", "OpenRouter",
"NVIDIA NIM", "Longcat", "Nara", "GLM/Zhipu"],
"fullstack": ["React", "Node.js", "FastAPI", "MongoDB", "PostgreSQL", "WebSockets"],
"devops": ["Docker", "GitHub Actions", "AWS", "Vercel", "Railway", "Render"],
"languages": ["Python", "JavaScript", "TypeScript", "C/C++"],
}
def flagship_projects(self):
return {
"BTSC-UNet-ViT": "Brain tumor classification + segmentation — hybrid ViT + U-Net on 90k+ images",
"DINOv2-ISIC": "Skin lesion classification — DINOv2 fine-tuned on ISIC 2019, achieved SOTA F1 Macro 0.7139",
"GPT-1": "GPT from scratch — nanoGPT trained on 35K WhatsApp messages, 523-token custom vocabulary",
"FinetuneLLM": "LLM fine-tuning pipeline — LoRA/QLoRA on custom datasets with FAISS-powered curation",
}
def current_mission(self):
return [
"🧬 Achieved SOTA in medical imaging — DINOv2-ISIC (F1 0.7139, beat ISIC 2019 leaderboard)",
"🧠 Trained GPT from scratch on custom WhatsApp dataset — nanoGPT, 523-token vocab",
"📦 Building LLM fine-tuning pipeline — LoRA/QLoRA with FAISS-powered data curation",
"🤖 Architecting multi-agent systems & full-stack AI developer tools",
]
|
Brain Tumor Classification + Segmentation Tech: PyTorch · ViT · U-Net · Medical Imaging research = {
"architecture": "Hybrid ViT + U-Net",
"dataset": "90,000+ medical images",
"tasks": ["Classification", "Segmentation"],
"approach": "Transformer-based deep learning"
} |
Skin Lesion Classification — SOTA Tech: PyTorch · DINOv2 · ISIC 2019 · Computer Vision result = {
"model": "DINOv2-B fine-tuned",
"dataset": "ISIC 2019 — 25k images, 8 classes",
"score": "F1 Macro 0.7139 — beat official leaderboard",
"status": "Shipped.."
} |
|
GPT From Scratch on Custom Data Tech: PyTorch · nanoGPT · NLP · Character-Level Tokenizer training = {
"base": "Karpathy nanoGPT — 6 layers, 4 heads",
"data": "35,213 WhatsApp messages",
"vocab": "523 tokens (8× Shakespeare's 65)",
"context": "256 tokens, character-level"
} |
LLM Fine-Tuning Pipeline Tech: Python · LoRA · QLoRA · HuggingFace · FAISS pipeline = {
"method": "LoRA/QLoRA parameter-efficient tuning",
"data_prep": "FAISS vector deduplication",
"routing": "Multi-model LLM ensemble",
"output": "Domain-adapted instruction models"
} |
