Mohamed Hassani

Architected and implemented scalable image generation platform for client exhibitions with over 200% improvement in visitors retention. • Developed full-stack solution including backend services, visitors info acquisition, GUI applications, and RESTful APIs. • Integrated computer vision for gesture control, enhancing user interaction. • Implemented cloud-based image delivery and processing pipeline. • Tech: Python, PyTorch, PyQt, OpenCV, FastAPI, RESTful APIs, QR codes, Cloud services, Git

• Created High-Quality Training Data: Generated precise and comprehensive training datasets to enhance the performance of LLM Agents models. • Reviewed Agents Outputs: Conducted thorough reviews of generated outputs to identify inaccuracies, improve error rates, and ensure the relevance of information. • Ensured Adherence to Annotation Guidelines: Strictly followed detailed annotation protocols to maintain consistency and accuracy in training data.

Led development of multiple computer vision projects from conception to production, with accuracy over 90% and few milliseconds responses. • Designed and deployed efficient high fidelity OCR and visual search systems with with accuracy over 90% and few milliseconds responses. • Developed and optimized data processing pipelines for large-scale image and text datasets for models training and evaluating. • Created RESTful APIs for model serving using FastAPI, Docker, AWS and Torchserve.

• Collection data on mental health discussions by scrapping from public websites. • Experimenting with extracting data from podcasts on mental health. • Building training datasets for different models. • Experimenting with different language models for generating natural responses. • Fine-tuning Hugging Face Conversational AI and DialoGPT for mental health support conversations. • Using rule-based Rasa model for conversation orchestration. • Deployment of the application as experimental API on GCP.