Hi, I'm Saif Mahmoud

High-fidelity content intelligence pipeline. I Intercepted internal network traffic to fetch raw metadata, then routed the videos through a custom VADER triage gate (sub-35ms latency). This filtering reduced GPU load by 60%, allowing the heavier Int8 Faster-Whisper model to run within a strict 190MB VRAM constraint. Authored a custom triton kernel to increase cold-start throughput by 71%, bypassing caching overhead and maintaining 0.96x sustained throughput

Bilingual SEO engine that drove a 288% traffic surge. I engineered a pipeline combining unsupervised K-Means clustering for niche identification and a perplexity-based validation gate (Sentence-Transformers) to act as an 'LLM-as-a-judge,' preventing low-quality content generation.