Senior AI Engineer — LLM systems, multi-agent architectures, edge AI.

From tensors → kernels → inference → production.

📍 Riyadh, KSA · 🟢 Open to consulting & advisory (part-time)

💰  >99.9%   Cut a fintech KYC pipeline from ~$100K/mo → ~$30/mo, fully on-prem
📹  30       Concurrent camera streams in a production PPE-detection system
⏱️  60–80ms  Edge inference on <4GB RAM hardware
🧠  70%      Lower fine-tuning cost via LoRA/QLoRA on 7B–13B models
👥  500+     Member community · 300+ active learners via SAiR

🏦 Senior Data / AI Engineer · Amlak International Finance Company · Riyadh, KSA · 2026 – Present Building production data pipelines and platform architecture for AI/ML initiatives — warehousing, transformation, orchestration.

🤖 Senior AI Engineer & LLM Consultant · Independent / Contract (Euro area) · 2023 – 2026 End-to-end ownership: architecture, evals, observability. → Redesigned a fintech KYC pipeline: ~$100K/mo → ~$30/mo (>99.9%), fully self-hosted, –60% manual processing. → LoRA/QLoRA fine-tuning on 7B–13B models: –70% cost · DeepSpeed/DDP/AMP: –20–30% training time · GGUF/ExLlamaV2: 3× faster inference. → Enterprise RAG over pgvector/Qdrant/Mongo (FastAPI): –40% query latency.

👁️ AI Engineer — Computer Vision · VisionCore · 2024 – 2025 → YOLOv8 PPE detection @ 85–90% mAP, 30 concurrent streams on 4×RTX 4070 via DeepStream. → TFLite/ONNX on <4GB RAM: 60–80ms end-to-end. –15% overfitting, –25% labeling effort.

🔬 R&D Engineer — AI & Deep Learning · Karabük University · 2021 – 2023 → LSTM-Transformer-CNN fusion for multimodal solar forecasting (NOAA GOES-16/17) — beat published RMSE/MAE/R² baselines (manuscript under review).

⚙️ Earlier — Industrial Automation & Renewable Energy · 2017 – 2020 PLC/HMI/SCADA (+18% uptime) · dual-axis solar tracker (+35% energy) · V2G microgrid optimization.

More: Kashif (tabular ML agent) · Solaris (multimodal solar forecasting) · PAPER2CODE · HandsOnLLMs

Free. Open-source. Built for engineers who have the curiosity but not the access.

I founded SAiR to build the AI engineering ecosystem I wished existed when I was starting out: serious AI education, research, and engineering for Sudanese and African engineers.

500+ community · 300+ active learners · 65% completion · 94% interview confidence · 3.2mo avg time-to-offer

6-module curriculum (Python/DS → ML → DL → GPT-from-scratch → MLOps). Flagship capstone miniGPT: configurable 10M–163M param architectures, from-scratch tokenizer/attention/training, DDP + Modal A100 training, fine-tuning modes, web UI, 39 tests.

Student teams now shipping in healthcare diagnostics, crop disease detection, Arabic NLP, LLM apps, and MLOps.

SAiR on GitHub · Join on Telegram

LLM / Inference: PyTorch Hugging Face vLLM Triton CUDA DeepSpeed LangChain RAG / Vector: pgvector Qdrant ChromaDB FAISS Vision / Edge: YOLOv8 ONNX TensorRT NVIDIA Jetson Data: ClickHouse dbt Airflow Infra / MLOps: Docker Kubernetes FastAPI Prometheus Grafana Languages: Python C++ TypeScript SQL Bash

• Senior Data/AI Engineer @ Amlak International Finance (Riyadh, KSA) — building data platform foundations for AI/ML.
• Going deep on inference engineering: quantization, kernels, GPU execution, serving (vLLM / SGLang).
• Extending MyTorch toward Triton/CUDA attention and matmul kernels.
• Growing SAiR's next cohort and open-source research track.
• 🟢 Available for select consulting & advisory on LLM systems, RAG architecture, and edge AI — book a consultation.

Understand the model, the runtime, and the hardware — not just the API call.

Languages: Arabic (native) · English (C1) · Turkish (B2) · Certifications