Evaluation of modern Retrieval-Augmented Generation (RAG) systems using generic and legal datasets. Thesis project for an Integrated Master’s in Electrical & Computer Engineering at the University of Thessaly.

Controlled depth ablation of a BERT bi-encoder across training budgets and seeds on three BEIR tasks (nfcorpus, scifact, fiqa). L3–L12 is flat within seed noise at 20K steps; 80K training degrades every depth on zero-shot transfer (−45% NDCG@10 on fiqa for L12).