$ whoami
$ role --current
_
BSAI student building computer vision models and LLM-orchestrated data systems that go past the notebook — trained, optimized, and shipped as working applications.
// about
I'm a BSAI student at the University of Central Punjab with hands-on experience training machine learning models, building LLM-orchestrated data systems, and optimizing data preprocessing pipelines. I'm drawn to the point where algorithmic theory turns into something a person can actually click on.
Currently working on my final year project — an explainable framework for Huntington's disease prediction — alongside computer vision and applied generative AI work.
// selected work
Instance segmentation pipeline using a custom-trained YOLOv8s-seg model on the DeepGlobe dataset to classify 6 land cover categories from satellite imagery. Custom OpenCV preprocessing converts pixel-wise RGB masks into normalized YOLO polygon coordinates. Tuned for 49.0 FPS inference on a Tesla T4 GPU.
Hybrid enterprise intelligence system pairing a distributed Apache Spark SQL engine with a Retrieval-Augmented Generation pipeline. A LangChain ReAct agent, powered by Groq's Llama 3.3 70B, autonomously routes natural language queries between the SQL and RAG engines, backed by an in-memory Qdrant vector store.
U-Net architectures with ResNet-50 backbones benchmarked across the DeepGlobe and LandCover.ai datasets, reaching peak mIoU scores of 0.637 and 0.798. Trained with discriminative learning rates, cosine annealing, and mixed-precision, then stress-tested across in-distribution, shift, and out-of-distribution scenarios.
End-to-end NLP pipeline automating text feature extraction and multi-class classification across 24 job sectors, reaching 72.23% cross-validated accuracy. Deployed with an interactive Gradio interface for real-time inference testing.
// contact
Open to internships and collaboration in ML and applied AI. The fastest way to reach me is email — I read everything.
abdullahahmed5u88@gmail.com