Introduction to the EVA Match AI Assistant

Our latest release introduces EVA AI Assistant in the Smart Match feature, a cutting-edge tool that leverages OpenAI's technology to enhance job-matching capabilities.

This guide will walk you through its functionalities and how to utilise them effectively.

Enabling EVA GPT Filter

The EVA GPT for Smart Match filter is turned off by default. Users can enable it by contacting their customer success manager.

Creating a EVA AI Assistant Match Filter

Access: via Search & Match > Jobs in the left menu or via the Jobs & Talent > Jobs > click on a Job and click on Match
Click on “Add new Filter” in the left menu in the Smart Match module
Select the “EVA- AI Assistant” from the “Edit Filter” drop down.

Customization: You can customise filters to get the desired results, including specifying search fields. For example, you can type full text in the “Request” field.
Application: Click 'apply' to activate the filter
Performance: The system waits up to 10 seconds for a reply from GPT and ElasticSearch

How to best leverage the EVA AI Assistant for Matching

EVA AI Assistant can interpret your requests in natural language. Below are a few examples demonstrating how the system facilitates search queries (also referred to as prompts) across different areas within your talent database.

You can tailor your queries for greater precision by specifying what to include and exclude.

Below are some examples:

Advanced Search Options

For advanced applications, utilise your Skill Taxonomies and proficiencies: This functionality enables searches based on specific skills or within particular skill taxonomies. For detailed guidance on incorporating Skills in the Match filter, please consult the following article

You can also leave feedback on this feature by clicking on the “Leave Feedback” button and adding your feedback in the comments section.

Please ask for more training on this feature from the customer success team.

Match GPT filter in Applications

The Match GPT Application Filter introduces an advanced text-based search capability within job pipelines to streamline application management and enhance workflow efficiency. Designed to support workflows with configured application states, this feature allows users to leverage AI-powered query building for precise and contextual searches. Users can create the "EVA - AI Assistant for Applications" filter through a simple interface by clicking "Add a new filter," selecting the filter type, and inputting their queries in the provided fields. This functionality enables additional searches for pipeline-related data, such as job screening questions or custom pipeline fields, ensuring comprehensive filtering options. For workflows with configured pipelines like ABC, the filter type appears dynamically, aligning with the application's setup. Upon filter creation, users are presented with an intuitive, empty text area to input queries, enhancing flexibility and usability while improving the overall recruitment process.

Match GPT > Query visualisation

The 'Match GPT Query Visualization' feature introduces an intuitive and user-friendly way to understand and manage MatchGPT filters in edit mode, enhancing transparency and control. When users interact with the 'Info' button for filters such as 'EVA - AI Assistant for Applications' or 'EVA - AI Assistant,' a modal is displayed. If the query request has not been applied previously, an error message is shown, providing immediate feedback. For applied requests, the modal reveals detailed request information, including the Elasticsearch query structure, using a back-end integration for accuracy. A code folding component ensures the query details are neatly organized, with all sections expanded by default for a complete view. This feature empowers users to easily review, understand, and troubleshoot MatchGPT queries, aligning with their workflow requirements and improving overall usability.

Example in the below screen shot shows filtering candidates by age group more than 20 and less than 60 yrs:

SMART MATCH GPT - Advanced

Overview:

SMART MATCH GPT enhances candidate search capabilities using natural language processing, allowing users to search for candidates based on various criteria such as custom fields, date of birth, nationality, language, and location. Additionally, the system utilizes RAG (Retrieval Augmented Generation) logic to analyze skills and match candidates based on specific work experiences and skill sets.

Supported Functionality:

Custom Field Search:
- Example Phrases:
  - Find candidates with Gender is Female
  - Find candidates with Internal Candidate Field is Yes
  - Find candidates with Company Relatives is No
Date of Birth:
- Example Phrases:
  - Find candidates under 65
  - Find candidates between ages 20 and 60
  - Find candidates born after 1960
Nationality:
- Example Phrases:
  - Find candidates with Nationality British or French
Language and Proficiency:
- Example Phrases:
  - Find candidates with Languages English, Spanish or French
Location:
- Example Phrases:
  - Find candidates with location is in London

WHED (World Higher Education Database) Recognition:

Recognized vs. Non-Recognized Universities:
The system allows users to filter candidates based on the recognition status of their educational institutions by the World Higher Education Database (WHED).
- Example Phrases:
  - Candidates have a WHED recognized University: This query will return candidates who have obtained their education from universities that are recognized by WHED.
  - Candidates have NOT a WHED recognized University: This query filters out candidates whose educational background is from institutions not recognized by WHED.

Education Degree:

Degree Level and Specific Fields:
Users can search for candidates based on their educational qualifications, including the degree level and specific fields of study.
- Example Phrases:
  - Find candidates with master degree or above in epidemiology, demography, statistics or health sciences: This query identifies candidates who hold a master’s degree or higher in specified fields, emphasizing their expertise in areas relevant to health sciences and research.
  - Find Candidates who have NO master degree or above in Public Policy, Public Health, Political Science, Public Administration, International Public Relations OR Law: This query excludes candidates who possess a master’s degree or higher in the listed disciplines, narrowing down the search to those with different educational backgrounds.

RAG Logic Overview:

Purpose:
RAG (Retrieval Augmented Generation) is used to extract and generate keywords from skill descriptions, which the system then utilizes to match candidates' work history with the search criteria provided by the user.

Example Scenario:

Skill: Parliamentary Office
Description:
A parliamentary office involves working within a legislative body to support lawmakers in developing, analyzing, and implementing public policy. Sub-skills include legislative drafting, policy analysis, constituent services, and political communication. Professionals assist in the creation of laws, conduct research to inform policy decisions, and engage with constituents to address their concerns. They often work directly with elected officials, committees, and other legislative staff. Proficiency in this field is typically achieved through education in political science, law, or public administration, coupled with experience in legislative internships or staff roles.
RAG Extracted Keywords:
"legislative drafting; policy analysis; constituent services; political communication; public policy development; law creation; policy research; elected officials support; legislative committees; legislative staff; political science; public administration; legislative internships; law analysis; policy implementation"
User Search Input in MatchGPT:
Find candidates who have experience in a parliamentary office
System Behavior:
The system uses the RAG-extracted keywords to search through the candidates' work history and matches those who have relevant experience as described by the extracted keywords.

Key Clarifications:

Lemmatization & Synonyms:
- Only lemmatization is applied; no external synonyms are used. Keywords are strictly derived from the skill description.
Parent Skill vs. Child Skill:
- If a search query references a Parent Skill (e.g., Political and Social Sciences), the system will aggregate keywords from all related sub-skills that are considered similar.
- If the query is for a Child Skill (e.g., Parliamentary Office), the system will only use keywords specific to that skill.
Similarity & Keyword Extraction:
- Similarity between sub-skills and keywords is determined using an LLM (Language Model) such as ChatGPT. There is no internal algorithm for this; decisions on similarity are made by the LLM.
Keyword Limitation:
- There is no specific limitation on the number of keywords the system can use. However, it focuses on the most relevant ones extracted from the skill descriptions.

Additional Supported Functionality:

Relevant Work Experience:
The system supports queries that focus on candidates' work experience, both in specific organizations and in particular roles or industries.
- Example Phrases:
  - Find candidates with experience in Company A or Company B: This query returns candidates who have worked with the Company A or the Company B, highlighting those with experience in international health and governance.
  - Find candidates with minimum of 3 years of experience in data analysis or data visualization: This search focuses on candidates who have at least three years of professional experience in data-related roles, which is crucial for roles that require analytical expertise.
  - Find Candidates who has NO minimum of 15 years of professional experience in external relations, international public relations or public policy: This query excludes candidates with significant long-term experience in external relations or related fields, potentially targeting candidates with less extensive but possibly more current or diverse experience.
  - At least two years of experience working in a parliamentary office: This query identifies candidates who have spent at least two years working in a parliamentary office, which is vital for roles requiring direct legislative experience.

These additional functionalities enable a more nuanced and targeted candidate search, allowing users to precisely filter candidates based on educational background, degree specificity, and relevant professional experience. This enhances the overall effectiveness of the SMART MATCH GPT system, ensuring that the most suitable candidates are identified for specific roles and requirements.

Final Notes:

The SMART MATCH GPT system is designed to be intuitive and precise, offering users a powerful tool to find candidates who meet specific criteria. For further clarifications, users can refer to the detailed product documentation or reach out to support.

EVA GPT in Smart Match Feature