ML-верификация, дедупликация и entity resolution
Entity Resolution: Canopy clustering + Siamese Network + XGBoost (precision 0.98, recall 0.95). Merging с provenance tracking.
Citation Verification (RAG): DeBERTa-v3 entailment (precision 0.97), source reputation weighting, multi-hop retrieval (ReAct).
Anomaly Detection: Isolation Forest + LSTM-AE (precision 0.92, recall 0.89). Data Quality Scoring (Great Expectations).
Stack: BGE-M3, Cohere Rerank, Llama-3.1-70B/GPT-4o, DeBERTa-v3, Qdrant, ClickHouse, Airflow.
← Назад в раздел

