issam jebnouni

► Built an AI-driven recruitment platform leveraging OpenAI embeddings, spaCy, and custom NLP pipelines to match CVs to job descriptions and detect LinkedIn inconsistencies. ► Designed a multi-stage analysis engine for resume parsing, job requirement extraction, similarity scoring, and automated match explanation generation. ► Developed and deployed the backend (Flask, Redis, Dramatiq) with production-grade infrastructure using Gunicorn, Nginx, and Cloudflare. ► Delivered the full-stack solution end-to-end, with a responsive Next.js 15 frontend.

► Trained a car damage segmentation model to identify 6 damage types in insurance subscription images. ► Created and containerized an API, enabling deployment and integration by the web team on Avidea’s demo page. ► Reduced the necessary time to process images by a factor of 3 and cut workforce requirements by 20%.

► Developed a Flask-based web interface for scanning and displaying local network RTSP camera streams. ► Explored the NVIDIA TAO toolkit 4.0 then trained and benchmarked various object detection models from its vast model zoo, achieving 0.90 mAP for the best model which was YOLOv4. ► Explored methods for enhancing model precision with limited real data, settling on training on mixed data (real + synthetic) and validating on real data only. ► Achieved a 0.96 mAP50 on the validation set using the YOLOv8 model with mixed data and generated a TensorRT engine to optimize the model for inference.