HAMZA AHMED SHARIF

代表作品 / 项目

• Project-6-DocMind-Agent⭐ 1Multi-agent RAG system with LangGraph — Researcher, Writer, and Supervisor agents with web search and document retrieval
• Project-5-PaperQA-RAG-System⭐ 1RAG system for PDF Q&A, built from scratch (no LangChain) — FAISS + BGE embeddings + FLAN-T5, with page-level citations and a relevance threshold to avoid answering from irrelevant content. Deployed on Hugging Face Spaces.
• Project-4.5-CIFAR-100-Image-Classification⭐ 1CIFAR-100 classification: EfficientNet-V2-S, WRN-28-10 & ViT+LoRA compared — LoRA wins at 92.49% test acc with ~1.5% trainable params
• Project-4---CIFAR-10-Image-Classification⭐ 13-stage CIFAR-10 classification pipeline: EDA, 3 model architectures (CNN from scratch, EfficientNet-V2-S, WRN-28-10), GPU benchmarking. Best: 95.85% test accuracy. PyTorch + mixed precision (AMP/FP16).
• Project-3---Customer-Churn-Prediction⭐ 1An end-to-end churn prediction system for ~982K customers. What makes it different: → XGBoost predicts churn with AUC 0.91 (82% accuracy, F1 0.68 at tuned threshold) → Hugging Face google/flan-t5-large generates human-readable explanations for individual predictions.