Idriss abderraouf Jedid

During my internship at Elyadata ,I designed a multi-agent system to generate AI/GenAI use cases tailored to different industries. I integrated LangGraph, LangChain, and LLMs to enhance reasoning and context-aware ranking, and built a RAG pipeline to ground outputs in real-world data. Additionally, I developed a FastAPI backend with an Angular frontend to deliver an interactive platform for exploring and visualizing AI-driven use cases

During my internship at Eagle Projects, I developed a robotic assistant with natural voice interaction using ROS 2 for controlling humanoid and quadruped robots. I implemented a multi-source ASR system combining Wav2Vec2, Whisper, and Vosk for robust speech recognition with fallback handling. I leveraged Mistral for grammar correction and reasoning over transcribed text, and integrated SpeechT5 for real-time speech synthesis to provide dynamic vocal feedback.

During my End-of-Year Project at Elyadata, I developed a FastAPI backend for automated extraction of invoice and receipt data. I trained a LayoutLMv3 model, achieving 99% accuracy and 97% F1 score on layout analysis and 85% accuracy on information extraction. Additionally, I experimented with prompt engineering using Qwen2 and Qwen2.5 to enhance generative extraction tasks.

During my internship at AVIDEA, I developed a .NET MAUI application for automated vehicle photo analysis in insurance workflows. I trained a ResNet model (3,000+ images) for car face detection achieving F1 score 0.9338, Accuracy 0.932, and a Vision Transformer model (4,000+ images) for classifying vehicle completeness with Accuracy 0.93. Both models were converted to ONNX and deployed for real-time mobile inference, enabling faster and more accurate insurance claim processing.