█████╗ ██╗ ██╗███╗ ███╗███████╗██████╗
██╔══██╗██║ ██║████╗ ████║██╔════╝██╔══██╗
███████║███████║██╔████╔██║█████╗ ██║ ██║
██╔══██║██╔══██║██║╚██╔╝██║██╔══╝ ██║ ██║
██║ ██║██║ ██║██║ ╚═╝ ██║███████╗██████╔╝
╚═╝ ╚═╝╚═╝ ╚═╝╚═╝ ╚═╝╚══════╝╚═════╝
class AhmedSaeedUmar:
role = "AI/ML Engineer & CS Undergraduate"
university = "Sukkur IBA University — BS Computer Science (AI)"
location = "Karachi, Pakistan"
interests = ["Computer Vision", "Deep Learning", "AI Search & Optimization", "Full-Stack AI Apps"]
currently = "Fine-tuning transformers and building full-stack AI systems"
fun_fact = "My AI made Karachi's transport network 60% more efficient 🚌"AI / ML
AI Engineering
Languages
Web & Frameworks
Databases & Tools
Graph-based route-finding over 20+ real Karachi locations using BFS, DFS, UCS, A*, and Simulated Annealing. Reduced nodes expanded by up to 60% vs UCS with custom admissible heuristics.
Python A* Simulated Annealing Graph Search Route Optimization
Fine-tuned ViT-Base/16 on ~6,800 weather images across 11 categories. Achieved 92.82% test accuracy and macro F1 of 0.93 with attention map visualizations.
Python PyTorch timm ViT Computer Vision Transfer Learning
Engineered a dependency-light RAG system that answers questions from documents using custom text chunking, vector embeddings, similarity search, and LLM context retrieval — built without high-level frameworks.
Python RAG LLMs NLP Vector Search AI Engineering
Full-stack face recognition app with Flask REST API + React frontend. Benchmarked Logistic Regression, Decision Tree, and KNN — hitting 89.7% accuracy on a custom dataset.
Python Flask React MobileNetV2 TFLite scikit-learn
Monkeytype-style typing test with live WPM/accuracy tracking, 3 timed modes, and randomized words — built in pure Vanilla JS.
HTML5 CSS3 Vanilla JavaScript REST API
- 🥈 2nd Place — SIBA TECH Speed Programming Competition (50+ teams) · Oct 2025
- 🛠️ Participant — SIBATHON Hackathon, Sukkur IBA University · Feb 2026
- 💼 10+ freelance projects delivered on Fiverr & Upwork

