Our Research & Resources

Research & Resources

Arche AI  machine learning models are rooted in the research, intelligence analysis, and data analytic work of co-founder Dr. Eric Bachura. Our models are trained on informed feature spaces that are carefully chosen and, in many cases, are the result of proprietary engineering processes. The output of our models sets us apart from standard assessment techniques with performance metrics, such as Accuracy and AUC measures, ranging from the mid 80’s to mid 90’s for classification models and MSE rates of 0.1 to 0.19 in Likert aligned regression models. However, the real value of our models and data-pipeline processes goes beyond the reliability of our output. Our procsses are able to operate on unobtrusively collected data and have an ingestation to analytic output turnaround that is far beyond that of traditional assessment processes. This allows our users to focus on interpreting the resuls and incorporating them into their own processes while simultaneously reducing the time it takes to provide the resulting analysis to their own stakeholders.

Dr. Bachura first tackled the problem of developing timely and reproducible psychological assessment processes as a PhD student. During workshop and training efforts he offered to faculty and fellow doctoral students, he was introduced to the research work of Dr. Michele Maasberg, whose dissertation work established a link between the Dark Triad trait of psychopathy and the proclivity for insider threat behavior. It was during this time that Dr. Bachura recognized the incongruity of the value this research work had with the nature of the underlying assessment. Setting out to identify informative sources of data available to organizations that had the characteristics of abundance and unobtrusive availability, Dr. Bachura identifed the value of spoken language as represented in job interviews, performance reviews, conferences calls, and similar settings.

During this time, Dr. Bachura was simultaneously working on a number of personal efforts that involved the collection of conference calls data. During a discussion of one of these efforts in a seminar, a fellow doctoral student from a different department requested assistance with his dissertation. Dr. Bachura utilized his data collected, analytics, and machine learning expertise in this collaboration. This ultimately resulted in peer-reviewed research exploring the impact of deceptive speech on financial analyst ratings. Dr. Bachura’s role in this project also saw the development of supporting assessment models (deception and veracity perception/suspicion). Additional collaborations exploring motivational drivers, phonetic convergence, and communication styles were also underway during this time. In all collaborations, Dr. Bachura took on the role of data collection, feature identification and engineering, and model development.