PCS

Conversational models and automated assessment are core to Spark and SimVox. The broader offering also includes physical simulators, examination activities, scenario libraries and training administration, which retain value apart from the generated patient dialogue.

Faculty define scenario checklists and can customize assessment criteria. Learners receive automatic feedback and can challenge credit decisions for faculty review. The system's confidence thresholds and conditions for escalating a problematic conversation are unspecified.

AI/Speech and AI/Assessment use custom language models trained with historical learner interviews. AI/Translate adds translation to an English trained core. The system moved to generative models by 2024, but base models and a production version history are unspecified.

PCS identifies its own AI/Speech and AI/Assessment model families and explains their custom training. Translation and speech synthesis form additional processing layers. External providers, complete subprocessor names and the terms governing each model route remain unspecified.

A 2023 Academic Medicine evaluation covered 491 Spark encounters in four Penn clerkships, with 313 confidence responses. Learner confidence improved across the year, but technical problems affected experience. This observational work predates the current generative models and does not establish improved clinical performance.

Service terms encrypt stored recordings and limit sharing without permission, while allowing product improvement and deidentified data reuse. Models use learner interview data for training. Retention has no fixed schedule, and no complete response process for harmful generated feedback is specified.

The service privacy policy reaches simulator recordings, patient scenario settings and account data. It identifies PCS as data controller, but does not establish business associate status, a BAA or an explicit prohibition on real patient information.

Service terms require encryption at rest and in transit, restrict employee access and promise notification of affected users within 72 hours of a detected breach. No independent security attestation or audit artifact accompanies those operational controls.

Spark, SimVox and ALEX support simulated clinical interviews, procedures and communication practice. Clinical decisions occur within training scenarios. The offering has no medical device claim for diagnosing or treating real patients.

The AI/Scenarios team supplies human feedback for model refinement, alongside stated fairness and inclusivity commitments. An earlier Penn evaluation found no significant task completion differences by race or gender. That result does not validate the current generative or multilingual models.

Credit Challenge lets learners dispute assessment decisions with faculty. That provides review within the educational program. Service terms disclaim warranties about results and do not attach a vendor remediation obligation to inaccurate generated feedback.

Spark supports two way connections with Canvas, Moodle, Blackboard and other LTI 1.3 learning platforms. These links exchange educational workflow data, rather than live patient records. A production clinical EHR connection is not established.

Spark is a cloud service accessed through screens or VR headsets, and SimVox hardware connects to PCS Cloud. The privacy policy allows international transfers without fixing a hosting region or specifying tenant isolation.

SimVox Lite is listed at $14,995 with five years of service and support, but the store marks it sold out. Spark uses a quoted price per student per year with minimums. Current purchasing therefore requires availability or contract confirmation.

PCS serves medical, nursing, emergency services and allied health education. Named users include Saint Louis University and Florida State University. Penn's earlier Spark evaluation covered four clerkships, supporting selected medical training settings rather than every advertised discipline.