Awais Khan

Associate AI Engineer - Generative AI & Image-to-Video Pipelines

Funsol Technologies • Tiempo completo

Jul 2025 - Present • 1 yr 3 mos

• Eliminated serverless cold start latency from 80 seconds to under 1 second by migrating model weight storage from Docker image layers to RunPod persistent volume mounts — directly reducing infrastructure costs for production GenAI pipelines. • Achieved 2x inference speedup on image-to-video pipelines serving 50,000+ users (Templix app) by integrating Sage Attention kernel optimizations and vmtouch memory caching to maximize GPU utilization. • Engineered automated CI/CD workflows using Docker and GitHub Actions for zero-downtime ML model deployments, enabling continuous iteration across production inference endpoints. • Designed and deployed end-to-end serverless inference pipeline for the ACE-Step 1.5 generative audio model, containerized with Docker and orchestrated to RunPod via GitHub Actions.