AI in Recruitment: Optimizing Candidate Search with RAG Chatbot

Customer Overview

Our client, a recruitment giant, followed a candidate search process that was time-consuming and cumbersome. It resulted in inefficiencies and low productivity, impacting their bottom line. They wanted to improve their services by optimizing candidate search.

Project Overview

The client needed a more efficient method to identify relevant resumes, surpassing the limitations of traditional filters and searches. They envisioned a Chat-based solution, where recruiters could query in natural language and quickly get information about candidates with required skills and experience.

Challenges

Understanding complex layouts of resumes to extract and categorize relevant data into skills, experience, etc., and accurately retrieving information from the index store to provide an appropriate response to the user’s question.

• The solution must handle multiple complex layouts, formats, and structures of numerous fancy resumes, it is difficult to parse them correctly.
• The information parsed and extracted from the resumes must be categorized and grouped accurately into skills, experience, qualifications, personal information, project details, professional certifications, and other applicable parameters.
• The solution must understand user questions and identify entities (like skills, qualifications, projects, experience, etc.), and using that information it must search/retrieve candidate information to feed into the LLM for response synthesis.
• Not restricted to merely candidate search, the solution must enhance other processes in recruitment like creating candidate profiles.

Solution

Combining our Vision AI and Generative AI expertise, we developed a Retrieval-Augmented Generation (RAG) solution integrated into a Chatbot to facilitate natural language-based candidate search.

• We used both Vision AI and Gen AI techniques to overcome the challenge of parsing diverse resume formats and complex layouts and structure the unstructured data of the resume document.
• We leveraged a Vision AI model that understands the resume document’s layout and detects different sections within that document. Then, using an OCR AI model, we extracted the text within these detected sections.
• Using a Gen AI model, we grouped the data extracted by OCR into categories like skills, project details, qualifications, personal information, etc., and created JSON for further processing.
• We indexed these JSON files per candidate into the OpenSearch vector store as both an embedding and normal text so that we could later perform Hybrid Search i.e. Vector plus Keyword-based Search.
• When a user queries the Chatbot, it understands the query, retrieves candidate information from OpenSearch using Hybrid Search & sends it to the LLM to synthesize the response based on the Question and Context.

Benefits

• The AI-powered Chat-based solution drastically reduces the time recruiters spend searching for relevant candidates.
• The system handles diverse resume formats and complex layouts. Provides accurate and contextual candidate information.
• With user-friendly interactions, improved search results, and data insights, recruiters make more informed and quick decisions.
• Benefiting from easy candidate search, recruiters focus on tasks needing human intervention, enhancing recruitment services.

Technology

• GPT 3.5
• OpenSearch
• Unstructured
• LlamaIndex
• Python
• FastAPI
• Angular

Industry

• Recruitment & HR