Abstract: Embodiments can relate to a system for detecting an anomaly, the system including a processing module. The processing module can extract an audio feature and a video feature. The processing module can generate an audio vector and a video vector. The processing module can amplify amplitude differences of spatial and temporal values between at least two video frames and/or phase differences of spatial and temporal values between at least two video frames. The processing module can perform a threshold comparison by determining whether an amplified amplitude difference is greater than a threshold amplitude difference and/or whether an amplified phase difference that is greater than a threshold phase difference. The processing module can determine change-in-position of an object associated with the extracted audio feature and corresponding video feature as a change-in-position anomaly or a change-in-position normality, and classify the data input as including a data manipulation or a no-data manipulation.

Abstract: Exemplary systems and methods conduct generate a plan for executing the anomaly detection operation without using or interacting with a secondary and/or external function or model. Any number of anomaly detection algorithms can be leveraged to generate a plan without human intervention or interaction. An LLM is trained to reason and autonomously identify anomalies in a dataset. The anomaly detection algorithms are arranged in a specified sequence to obtain a solution. Once the plan has been successfully executed, a self-reflection operation is performed to identify an flaws in the plan based on the goal or task. The plan is revised to mitigate any identified flaws. After one or more iterations of self-reflection and plan revision, a cohesive plan is obtained and is executed without errors and with successful anomaly detection/identification in the dataset.