Applied AI Engineer & CS Student | Cairo, Egypt

Applied AI Engineer building LLM agent pipelines integrated with CRM systems. Previously API Engineering Intern at GCP working on Apigee X REST/SOAP APIs.

LinkedIn | Email | Hugging Face Profile

PyTorch, OpenCV, SciPy, LangGraph, ChromaDB, SQLite, Docker, GCP, Streamlit

GitHub Repo | Live Demo

• Architecture: Fuses ConvNeXt-Small spatial embeddings with SRM + Bayar-Stamm 2D Real FFT spectral embeddings via a sigmoid residual gate.
• Identity Splitting: Graph connected-component partitioning (networkx.Graph) over actor IDs to enforce 0% identity overlap across train/val/test splits.
• Calibration & Interpretability: SciPy L-BFGS-B temperature scaling ($T^*=1.4788$, ECE = 0.0093) paired with an on-demand 4-panel Grad-CAM diagnostic engine.
• Metrics & Testing: 0.9987 test AUC, 0.9830 F1-score, verified by a 54-test pytest suite.

GitHub Repo | Live Demo

• Two-Stage Retrieval Engine: Combines ChromaDB vector search with a custom PyTorch CarrierReRanker MLP model for carrier dispatch ranking.
• LangGraph Multi-Tool Agent: Autonomous tool-calling agent routing between SQL queries, semantic vector search, and density-based NMFC freight class calculations.
• Guardrails & Resilience: Connection-level read-only SQLite isolation, SELECT-only SQL validation, and exponential backoff retry wrappers for LLM rate limits.

• Fixed documentation bug miscategorizing SmolLM3 under Multimodal models in _toctree.yml. Merged into huggingface:main.