.svg)
FedTuna
About the Organisation
FedTuna gives AI/ML engineers data intelligence for models running on private, regulated data. Its agents run inside the customers infrastructure, on-prem or private cloud, with no GPU requirement, and data never leaves the secure perimeter. When a models accuracy drops, FedTuna explains why. It analyses model performance against the specific production data, showing which data segments, sources, devices or input conditions the model handles well and where it fails. This identifies the cause of degradation (data drift, new data patterns, under-represented cases) and tells engineers what to do next: adjust thresholds, collect targeted data, or retrain. Findings are delivered as reports for engineering, model risk and compliance teams. Where retraining is needed, FedTuna can fine-tune the model on fresh local data using privacy-enhancing technologies and federated learning; only model updates move. Supports in-house, vendor and frontier models, computer vision and tabular.
Products and Services
-
FedTuna
FedTuna is an agentic platform that keeps AI models accurate after deployment in regulated environments where data cannot leave the secure perimeter. Agents run inside the customers infrastructure, on-prem or private cloud, with no GPU requirement. - Monitor: track model performance and data drift on live production data, across in-house, vendor and frontier models. - Diagnose: pinpoint which inputs, segments or data sources are driving degradation. - Report: produce performance and drift reports for model risk, compliance and regulatory review. - Fine-tune: retrain models on fresh local data using privacy-enhancing technologies and federated learning. Only model updates move; customer data never does. Built for banks, fintechs and the AI vendors serving them, including fraud, AML, KYC and biometric models. Supports computer vision and tabular models.
