top of page

The science of Neufast Patent-pending Debiased AI

Our mission is to provide you and your candidates with engaging and highly accurate recruitment without bias. We ensure our scores are always valid, reliable, and fair by our deeply scientific, multi-layered Debiased AI/ML model.

Visual, Audio and Textual Based Analysis

Candidate’s apparent features, prosody features and semantic features are extracted from his/her videos based on computer vision, speech recognition and natural language processing techniques.

Visual channel for apparent feature extraction

Visual channel for apparent feature extraction

Apparent features such as facial expression and gesture are implicitly extracted by computer vision technique.

cloud computing

Audio channel for prosody feature extraction

Prosody features such as tonality, pace and rhythm are implicitly extracted by audio analysis technique

Semantic feature extraction

Textual channel for semantic feature extraction

The syntactic and semantic meaning of candidate’s spoken texts are extracted by natural language processing technique with contextual awareness

Neufast Patent-pending Debiased AI
Model Architecture

Our Patent-pending Debiased AI model analyzes the audiovisual and semantic features in a comprehensive way with complex task-oriented adaptive weighting scheme to infer candidate’s Impression Score and Potential Work Performance Score.

AI model architecture

Training of Debiased AI Model

Video Interview

Visual, Audio and Textual channels

(Deep Learning Models)

Apparent Features



Prosody Features



Semantic Features
Customized Regression

Customized Regression

Impression / Competency Scores

Impression / Competency Scores



Internal Governance of Developing Ethical AI Model

Neufast embraces Ethical AI and data ethics in the operation, the development and the application of AI following Ethics Guidelines for Trustworthy AI published by the European Commission in December 2018 [1]. To ensure the ethical development and use of AI in Recruiting, the data annotation and AI model development process of Neufast follows the 7 Ethical Principles of AI for internal data governance: 1. Accountability, 2. Human Oversight, 3. Transparency and Interpretability, 4. Data Privacy, 5. Fairness, 6. Beneficial AI, and 7. Reliability, Robustness and Security [2]. Decisions made by Neufast's patent-pending Debiased AI are EXPLAINABLE, TRANSPARENT & FAIR as Personal Identifiable Information (PII) are redacted. Neufast's patent-pending Debiased AI systems are designed to be HUMAN-CENTRIC [3].

Data Preparation
  • No manipulation of the training data collection leading to a biased model in favor of some certain group

  • No data preprocessing that changes the data distributions for different groups

​Data Preparation

AI Model
  • All key aspects of the model building process are reviewed

  • Salient numerical results are copied into the sheet

  • All significant deviations from the standard model training process are well documented

Model Building &

AI model testing
  • Adverse Impact Assessment is done to evaluate if our trained AI model complies with the UGESP's four-fifth rule* with respect to different groups in either: Demography, Educational background or other groups defined by clients

Model Testing

*Reference: Equal Employment Opportunity Commission, Civil Service Commission, et al. 1978. Uniform guidelines on employee selection procedures. Federal Register 43, 166 (1978), 38290–38315.

​Assessment of Impression performance

Neufast’s patent-pending Debiased AI video assessment tool assesses candidate’s capability in making a positive impression on others through three perspectives:



Positive emotions

​Assessment of Motivation

Neufast’s Debiased AI video assessment tool identifies candidate's motivation to initiate work-related behaviors for person-job fit & talent retention.

​Assessment of Motivation

Assessment of
Work Performance

Neufast’s competency model is built from cumulated studies on work performance. Candidates will be assessed through their task, contextual and adaptive performance.

Comprises job-specific behaviors which include accomplishing the role responsibilities and activities listed in in job description.

Task Performance

Job and Role Responsibilities

Leadership and Management Behaviors


Job and Role Responsibilities

Leadership and Management Behaviors

Job and Role Responsibilities

Leadership and Management Behaviors

Job and Role Responsibilities

Leadership and Management Behaviors

Contextual Performance

Comprises non-job specific behaviors that help maintain or improve the broader organizational, social or psychological environment necessary to facilitate task performance and achieving organizational goal accomplishments

Relationship and Communication

Contextual performance

Comprises behaviors and attitude required to be flexible and be able to quickly adapt to dynamic work situations

Adaptive Performance

Adaptation and Change

Adaptive performance

Reliability study

Our AI video assessment tool is highly reliable and accurate to our client’s internal rating. We analyzed the efficiency of our product by comparing our AI scores to HR (Client) internal rating: 

Cronbach’s Alpha > 0.7 (out of 1) for reliability​

0.3 - 0.6 (out of 1) positive Pearson correlation which is highly correlated to HR candidate ratings at p<0.05 significant level at competency-level.

F1 classification accuracy reaches up to 98% at competency-level.

Reliablity study


1. Ethics Guidelines for Trustworthy AI, High Level Expert Group on Artificial Intelligence (HLEG), European Commission, December 2018.

2. Guidance on the Ethical Development and Use of Artificial Intelligence, Office of the Privacy Commissioner for Personal Data, Hong Kong.

3. Neufast Limited United States Patent and Trademark Office Application Number 63/335,756; Confirmation No. 7481. 


bottom of page