Interview scoring utilizes six advanced algorithms, some customizable during interview setup, and others leveraging pre-defined scoring bands for consistent, insightful results.

Interview creation

In JobMojito, interviews can be created in two ways:

1. Automatically using a large language model. Based on the job title, job description, and additional metadata, the platform generates an

   1. interview sequence, a set of questions that will be asked by the avatar during the interview. These include test questions to ensure the candidate has at least a basic understanding of the field they’re applying for

   2. Scoring rubric, which are hidden criteria outlining what is expected from the candidate’s profile

   3. Interview Short & Long Descriptions

2. Manually, where recruiters define both the interview sequence and candidate expectations themselves.

Scoring Rubric

AI full interview analysis and AI interview question/answer evaluations are scored using a "scoring rubric." The scoring rubric is a set of expectations defined for each score bracket: Strong, Moderate, and Weak.

Each scoring bracket contains multiple requirements.

Recommendations:

1. Provide 3–5 requirements for each scoring bracket.

2. Do not repeat requirements between brackets.

3. Each requirement should be stated positively. Clearly describe what the candidate should do or demonstrate to fulfill the specific requirement. For example: "Candidate should have at least a bachelor's degree." Avoid negative wording such as: "Candidate did not finish university."

After each interview is completed, individual results are matched against the defined scoring rubric. Requirements are marked as follows:

• Green checkbox: Specific requirement was confirmed during the interview

• Red checkbox: Specific requirement was rejected during the interview

• Gray line: Requirement was not discussed during the interview

Customizing scoring weights

Before publishing the interview, recruiters can customize the scoring weights for each algorithm that JobMojito runs. If a score is set to 0, that specific algorithm will not be executed for any candidate in the interview.

Variable Descriptions

• Maximum Number of Attempts: Allows candidates to retake the interview multiple times.

• Disable Retry When User Reaches a Score: When enabled with multiple attempts, this feature prevents further retries once the candidate achieves a specified score—helping candidates avoid unnecessary attempts for minimal score improvements.

• Weights of Each Score Relative to Overall Score: Determines how much each individual score contributes to the overall result, letting you fine-tune the impact of each evaluated area.

Scoring Calculation

This section is describing, how individual scores are calculated

Score breakdown

You can see on the result page, how individual scores has been calculated and how the total score was determined.

AI analysis of full conversation

Is LLM based scoring method that is executed, once full interview is completed

Input:

• Position name

• Position description

• Candidate expectations

• Full transcript of the interview

Output is score between 1 - 10 where

• Score 1: for candidates that you would never hire based on their responses

• Score 5: for average candidates or candidates that didn’t completed the full interview

• Score 10: for candidates you would hire on the spot without any further testing

Purpose: To accurately assess how well the candidate aligns with the requirements of the job position.

You can access raw AI reasoning in the result page to better understand, how the score has been created

AI analysis of individual answers

LLM-based scoring is applied to each individual question and answer, ensuring precise evaluation of candidate responses.

Input:

• Position name

• Position description

• Candidate expectations

• Avatar question

• Candidate response

Output is score between 1-5 (translated into score between 1-10)

• Score 1: The answer missed the point

• Score 2: Partially addresses the question but lacks key details

• Score 3: Adequate and correct

• Score 4: A strong answer that goes beyond the basic requirements of the question

• Score 5: Demonstrates expertise, creativity, or exceptional depth

Purpose: To determine whether the candidate is answering each question accurately.

Speech sentiment

This is a text-based assessment of the candidate’s answer, conducted for each individual response.

Input:

• Text of the candidate answer

• Answer language code

Output:

• Sentiment between -1 and 1

• Score 0: When sentiment is -0.2 or lower

• Score 10: When sentiment is 0.3 and higher

Purpose: To detect whether the candidate uses profanity or provides generally negative responses.

External vendor: https://learn.microsoft.com/en-us/azure/ai-services/language-service/sentiment-opinion-mining/overview

Speech pronunciation

Audio analysis compares the candidate’s raw audio input against a native speaker’s voice profile. This analysis is performed for each of the candidate’s answers.

Input:

• Candidate response audio data

• Answer language code

Output:

• AccuracyScore: Pronunciation accuracy of the speech. Accuracy indicates how closely the phonemes match a native speaker's pronunciation. Syllable, word, and full text accuracy scores are aggregated from the phoneme-level accuracy score, and refined with assessment objectives.

• FluencyScore: Fluency of the given speech. Fluency indicates how closely the speech matches a native speaker's use of silent breaks between words.

• CompletenessScore: Completeness of the speech, calculated by the ratio of pronounced words to the input reference text.

• ProsodyScore: Prosody of the given speech. Prosody indicates how natural the given speech is, including stress, intonation, speaking speed, and rhythm.

• Score: Overall score of the pronunciation quality of the given speech. Is calculated from AccuracyScore, FluencyScore, CompletenessScore, and ProsodyScore with weight, provided that ProsodyScore and CompletenessScore are available.

Purpose: To assess the candidate’s spoken language proficiency in comparison to a native speaker.

External vendor: https://learn.microsoft.com/en-us/azure/ai-services/speech-service/how-to-pronunciation-assessment?pivots=programming-language-javascript

Speech cadence

During speech-to-text transcription, the platform records the exact millisecond each word is spoken. This data is converted into an industry-standard words-per-minute score, calculated using a purely mathematical formula.

Input:

• Words per minute data

Output:

• Score 1: When words per minute is less than 80 or more than 220

• Score 10: When words per minute is between 100-130

Purpose: To evaluate the candidate’s ability to produce naturally flowing speech.

Human score adjustment

On the results page, recruiters can provide feedback and increase or decrease the candidate's overall score. This traceable approach ensures a fair and transparent selection process.

Risk assessment

Automated risk assessment (must be enabled) evaluates interview risks and may reduce the score. See the risk assessment document for details on how the individual checks work.