yomna waleed Hassan

Contributed to AutonomIQ, a Hybrid RAG system for querying AUTOSAR and ASPICE PDFs; participated in framework evaluation (PathRAG, GraphRAG, HybridRAG, VanillaRAG) and recommended HybridRAG for production. Built a PDF parsing pipeline (initially Azure OpenAI API; migrated to Docling 2.63/2.64) to extract text and insert image placeholders (). Implemented image extraction and pre-processing: separated logos using OpenCLIP LION 2B, classified UML diagrams with YOLOv8-small into Activity, Sequence, Component, StateMachine, and Other (94% detection accuracy). Designed image-specific prompt templates and integrated Ollama QWEN 3VL for automated UML parsing; reinserted structured JSON outputs into original Markdown for accurate context. Implemented LangChain-based chunking (heading-aware), embedding pipelines, and ChromaDB indexing to enable context-aware retrieval; integrated GPT-OSS via Ollama for final answer generation. Automated Markdown report generation, reducing manual documentation time by ~60% and improving retrieval relevance by ~35%.