Ammar Daruger

Built an adaptive video conversion scoring platform (FastAPI service) that scores source video quality using OCR, ASR, face detection, and YOLO tracking. Implemented auto-cropping into multiple social-media variants with dynamic cropping, language dubbing, and subtitle generation. Developed a vehicle classification and damage detection system for a Saudi Arabian market client using a custom CUDA-trained model to classify car make/model, detect damage, and generate repair cost estimates. Delivered as a FastAPI service for integration in downstream applications.

Delivered an adaptive video conversion scoring platform via FastAPI. Scored source video quality using OCR, ASR, face detection, and YOLO tracking, then auto-cropped into multiple social-media variants with dynamic cropping, language dubbing, and subtitle generation. Built a vehicle classification and damage detection system for a Saudi Arabian market client using a custom CUDA-trained model to classify make/model, detect damage, and generate repair-cost estimates, served via FastAPI.

Enabled scalable AI deployment by designing modular Python backend services, reducing integration effort by ~30%. Improved NLP accuracy by fine-tuning classification and sentiment models (targeting ~15% gains) and grounding LLM outputs in domain data using RAG components. Containerized services with Docker/Kubernetes and automated CI/CD (build-to-release in days to hours). Shipped REST APIs integrating model-to-application workflows.