Rahhal EL BOUDALI

Designed and delivered an AI-powered legal document analysis system enabling automated compliance checks, risk detection, and structured insights from unstructured legal texts. Key AI & Technical Contributions: – Built an end-to-end Retrieval-Augmented Generation (RAG) pipeline for legal document analysis (PDF, DOCX, TXT). – Implemented robust document ingestion & text extraction using pdfplumber, python-docx, and native text loaders. – Designed a token-aware chunking strategy (512 tokens with overlap) to preserve legal context and improve retrieval accuracy. – Generated semantic embeddings using Sentence Transformers (all-MiniLM-L6-v2) for efficient document understanding. – Deployed a vector database layer using PostgreSQL + pgvector to store embeddings with metadata (page numbers, sections). – Implemented semantic similarity search to retrieve the most relevant legal clauses at query time. – Integrated LLM inference (LLAMA 3.1 – 8B) to perform contextual legal reasoning over retrieved document sections. – Engineered structured AI outputs (JSON) including: – Compliance scoring – Legal risk & red-flag detection – Actionable recommendations – Key entity and clause extraction – Focused on explainability and traceability, returning cited sections and page references alongside AI results. – Designed the system to be scalable, model-agnostic, and production-ready.

Worked on RoozCar, an AI-powered social automotive marketplace, delivering production-ready machine learning models and cloud data pipelines. Key AI & Data Contributions: – Designed and implemented an end-to-end data pipeline (PostgreSQL → BigQuery) using an ELT architecture for large-scale analytics. – Built a vehicle price prediction system using BigQuery ML, evaluating multiple models (Linear Regression, DNN, AutoML, Boosted Trees) and selecting the best-performing Boosted Tree model (R² ≈ 0.63). – Developed a hybrid recommendation engine combining: – Collaborative filtering (custom Matrix Factorization in SQL) – Popularity-based and interest-based heuristics – Improved content discovery and user engagement. – Performed advanced data preprocessing: – Feature engineering – Class balancing (oversampling) – Outlier detection using IQR – Automated model training, evaluation, and inference directly in BigQuery for cost-efficient AI deployment. – Created analytics dashboards & KPIs using a star schema data model to support business decision-making. – Exposed ML models via REST APIs for real-time predictions in the web application. – Collaborated closely with full-stack teams to integrate AI features into production.