Ayaan Sharif

• Produced and delivered AI-generated short-form promotional visuals for singer Nakash Aziz, building repeatable ComfyUI, higgsfield and other open source and proprietary tools, and built
workflows for image-to-video generation and cross-scene reference consistency.

• Built a Python-based processing pipeline for a 13, 000-film enterprise catalogue, ingesting 4, 000 available cloud assets and
processing an initial batch of 1, 000+ feature-length movies for compliance and metadata-analysis use cases.
• Integrated shot-boundary detection, event stitching and video splitting to transform long-form movies into model-ready clips
for speech, vision and multimodal inference workflows.
• Developed pipeline orchestration for GCS downloads, clip processing, Vertex AI inference, result aggregation, GCS uploads
and local cleanup, with per-movie state tracking for resuming interrupted processing stages.
• Implemented parallel clip-level compliance analysis using Gemini 2.0 Flash through Vertex AI, including configurable
concurrency, retry handling and structured JSON/JSONL output validation.
• Added GPU availability checks, CUDA-specific failure handling, disk-usage guardrails and Linux background-execution
scripts to improve the reliability of long-running processing jobs.
• Built an internal text-to-short-video POC using FastAPI and LangGraph, orchestrating prompt-to-story generation, scene
planning, image generation and video generation through conditional tool calls, with PostgreSQL-backed sessions and SSE
progress streaming.