AI/ML Engineer & LLM Expert specializing in Large Language Models, Generative AI, RAG pipelines, and Intelligent Automation. I build AI systems from the ground up — from GPT models trained from scratch to production-ready FastAPI deployments.
I'm a Computer Science graduate from FAST National University of Computer and Emerging Sciences (FAST NUCES), Islamabad, with a deep passion for Artificial Intelligence, Large Language Models, and Generative AI.
What sets me apart is that I don't just use AI frameworks — I build from scratch. I've implemented GPT and Transformer architectures from the ground up in PyTorch, fine-tuned LLMs on domain-specific corpora, and shipped production-grade AI systems end-to-end.
My work spans LLMs, NLP, Computer Vision, RAG systems, Knowledge Graphs, and Intelligent Automation. I'm currently executing a structured AI/LLM engineering roadmap, actively working with LangChain, LangGraph, agentic workflows, and voice AI.
Guided 30+ students in Operating Systems & Intro to CS. Evaluated assignments, conducted debugging sessions and exam preparation.
A research study evaluating quantum random number generators (QRNGs) using IBM's Qiskit Aer simulation framework. The research focuses on assessing the statistical quality, performance benchmarks, and noise-resilience properties of QRNGs across various quantum circuit configurations, leveraging Shannon Entropy metrics and noise mitigation techniques.
Technologies and tools I work with to build intelligent AI systems.
Real systems built from the ground up — not tutorials, not demos.
Production-grade RAG system over 5000+ Pakistani legal documents. Fine-tuned an SLM with LoRA/QLoRA, built LangChain agent with knowledge graph integration, deployed via FastAPI — end-to-end from data to inference.
A decoder-only GPT (~10.8M parameters) built entirely from scratch in PyTorch — no pretrained weights, no external tokenizers. Trained on Urdu poetry corpus from Rekhta. Generates coherent Urdu poetry from a single start token.
Complete encoder-decoder Transformer implemented in PyTorch from first principles — multi-head self-attention, sinusoidal positional encoding, LayerNorm, and residual connections. Every design decision documented with reasoning.
CNN-based binary classifier for handwritten signature verification. Led a 10-member engineering team through the complete ML lifecycle — data preprocessing, model training, hyperparameter tuning, evaluation, and delivery.
Open to opportunities, collaborations, and interesting AI projects.
Whether you have an exciting AI role, a project idea, or just want to talk about LLMs — I'd love to hear from you.