Applied AI Engineer & CS Student — Cairo, Egypt
Building agentic AI systems and retrieval pipelines. Previously API Engineering Intern at Cloudypedia (GCP Apigee X, REST/SOAP).
→ Full profile & experience on LinkedIn
| Area | Tools |
|---|---|
| ML / CV | PyTorch, OpenCV, SciPy |
| Agents / Retrieval | LangGraph, ChromaDB, SQLite |
| Infra | Docker, GCP, Streamlit |
Fuses ConvNeXt-Small spatial embeddings with SRM/Bayar-Stamm 2D Real FFT spectral embeddings and a Bi-GRU video sequence head. Identity-disjoint splits via graph connected-component partitioning (zero actor overlap). Calibrated via SciPy L-BFGS-B temperature scaling (T*=4.288, ECE 0.0785). 0.8719 AUC, 0.8818 F1 on a 13,444-crop held-out test set — 142-test pytest suite.
LangGraph agent routing freight carrier queries between exact SQL, hybrid retrieval (SQLite FTS5 + ChromaDB via Reciprocal Rank Fusion, cross-encoder reranked), and live FMCSA safety verification. Hybrid reranking lifts Recall@1 0.30→0.70, Recall@5 0.667→0.867, MRR 0.429→0.764 over dense-only retrieval on a 60-query benchmark — though plain SQL still wins on structured queries.
huggingface/transformers #46770 — fixed a doc bug miscategorizing SmolLM3 under Multimodal models.
Writing up the deepfake detector's leakage-fix and generalization findings for an arXiv preprint.
Email / Hugging Face in LinkedIn contact info.