Aditya Handrale

Short-term founding engineering contract to build an AI-powered image-to-video product for real-estate marketing. Architected an LLM-driven “AI Director” to generate storyboards, prompts, pacing, text overlays, and music selection, translating listing data into controllable video generation plans. Designed a storyboard-first UX to configure shots and trigger batch image-to-video generation with a predictable output structure. Benchmarked multiple image-to-video models across temporal consistency, motion coherence, and cost; selected production defaults based on quality–latency tradeoffs. Built async generation pipelines using Django APIs and Next.js frontend to support concurrent jobs and failure recovery. Implemented AI-assisted image cleanup to improve downstream video quality for demo assets used in investor discussions.

Sole technical founder who architected and deployed the AI and product stack for a generative fashion SaaS platform (Virtual Try-On, AI imagery), serving brands like Vogue and MCM and generating C30,000+ in early revenue. Architected end-to-end pipelines for AI fashion imagery and virtual try-on, covering data ingestion, inference orchestration, and user-facing workflows. Designed a domain-specific creative assistant ("Luca") using VLM/LLM components to generate campaign plans, text-to-image prompts, and SEO-optimised PDP copy. Built DreamBooth/LoRA training workflows to create photorealistic digital twins, balancing quality, inference cost, and deployment complexity. Deployed GPU-backed inference pipelines on Modal and RunPod with a FastAPI backend, Supabase-based auth/storage, and CI/CD pipelines to support iterative releases and client delivery.

Developed a visual analytics platform to evaluate high-dimensional feature spaces from industrial QA image datasets, supporting model inspection and failure analysis. Implemented and benchmarked clustering metrics and dimensionality reduction techniques (PCA, t-SNE) to detect outliers, mislabels, and robustness issues in defect detection classifiers.