RAGless: Q-Q retrieval with score aggregation for closed-domain FAQ [P]

What it does RAGless is a semantic retrieval system based on Question-to-Question matching. At ingestion, an LLM generates multiple question variants per answer (3–5) and each variant gets its own embedding. At query time, the user question is embedded, Top-K nearest question variants are retrieved, and scores are aggregated by answer_id — the answer with the highest aggregated score wins. Threshold logic uses two gates: minimum aggregated score (default 0.70) plus a fallback on the best single-hit score (0