Walid Ghenaiet

I tackled the challenge of data scarcity in automated visual quality control by designing generative pipelines capable of producing realistic and diverse synthetic defects. This data allowed for the training of supervised models that achieved performance comparable to, or even exceeding, state-of-the-art unsupervised methods.

Designed an intelligent agent capable of transforming simple questions into professional financial reports, significantly reducing research and writing time. The system utilizes an LLM (Gemini) to enrich queries, extracts relevant information from online sources (DuckDuckGo API) and a vector database (Pinecone), and then synthesizes the results into structured content.

Before export, the agent verifies the report's completeness: if the answer is deemed insufficient, it autonomously continues collecting and reformulating data to ensure accuracy and exhaustiveness.

This project illustrates the potential of autonomous AI agents to automate research, optimize financial analysis, and produce reliable, ready-to-use reports.

Designed and implemented a comprehensive machine learning system capable of translating raw electromyographic (EMG) signals into real-time commands for a Unity application. The project covers the entire pipeline, from signal acquisition and processing to model deployment and inter-process communication.

* A Gradient Boosting Classifier for multi-class gesture recognition (rest, fist, open hand), achieving over 92% accuracy.

* K-Means clustering for unsupervised fatigue level detection (in shape, mild fatigue, fatigue).