Danya Labasi

Conversational AI with Retrieval-Augmented Generation (RAG) and Context-Aware Generation (CAG): developed RAG systems to allow staff to query internal policies; integrated CAG to reduce latency and improve accuracy for static banking content. Designed multi-agent architecture capable of processing large documents in parallel using specialized AI agents with final aggregation logic, leveraging semantic chunking and token-efficient overlaps. Built modular preprocessing and training pipelines using spaCy, scikit-learn, and Hugging Face Transformers for scalable experimentation. Achieved an F1 score of ~0.92 on conflict detection and 100% accuracy on neutral test cases, outperforming baselines. Document intelligence: built scalable calable parsing pipelines using LlamaParse, OCR engines, and NLP; integrated data extraction from PDFs, images, and bilingual (Arabic–English) content, including non-OCR documents. Platform integration & support: managed UAT, VPN, and API integration tasks acr