Suki AI

AI hallucinations are instances where the AI generates inaccurate or fabricated information. Suki minimizes the occurrence of hallucinations by employing a meticulously designed prompt structure and robust grounding techniques. These techniques tie AI-generated content to factual information from the EHR and the encounter transcript, enhancing accuracy and reliability. Clinician review of all AI-generated suggestions serves as a primary safeguard User trust and safety are top priorities at Suki, and we continuously work to address potential bias and fairness. We identify potential bias sources in our ASR training data and work to mitigate them through targeted data augmentation and model retraining. For our ASR model, we incorporate a wide variety of user accents in our training data. We do not label or use sensitive PII like race or ethnicity in the training of our models or in clinical audits of output quality.