project-proposal-2025

Recruit.ai

Abstract

Imagine you’ve just landed a Software Engineering interview at your dream company. Instead of immediately turning to LeetCode, what if you had a personalised strategy? Recruit.ai offers a focused approach by analysing your unique coding skills, projects, and previous roles, alongside the specific requirements of the job description and applied company. This allows you to craft an interview simulation that mirrors the real challenges you’ll face. You’ll receive targeted feedback on your technical abilities and behavioral responses, ensuring you’re prepared to articulate your value and excel in the interview. With Recruit.ai, you move beyond generic preparation and gain a competitive edge by showcasing your unique strengths.

Author

Name: Saakshi Gupta

Student number: 48246006

Functionality

Recruit.ai offers a comprehensive system:

1. Dynamic Knowledge Graph (RAG): Gathers job descriptions, company data, resumes, and previous questions from diverse sources (LinkedIn, company sites). Extracts skills, requirements, identifies skill gaps. Connects candidates, companies, roles dynamically, updating with trends.

2. Personalised Simulation (LangChain): Generates adaptive questions, tailoring format and style based on the candidate’s profile, target role, company culture, identified skill gaps, and past performance. Scenario questions mimic real-world challenges relevant to the industry.

3. Multi-Agent Collaboration (LangGraph): Simulates varied interviewer personalities (technical expert, HR manager, challenging interviewer). Creates negotiation, conflict resolution, and stress-inducing scenarios. Provides holistic feedback from agent perspectives adapting their styles.

4. Advanced Performance Analysis: Uses NLP (sentiment analysis, topic modeling, accuracy assessment) to analyse responses. Pinpoints strengths/weaknesses in communication, technical knowledge, problem-solving, and cultural fit.

5. MCP Extensibility: Employs modular architecture for new interview types (coding, system design), third-party tool integration, customisable metrics, adaptable reporting.

Scope

The MVP delivers core personalised simulation:

1. Enhanced RAG: Ingests job descriptions, resumes, basic company overviews, and prior questions; extracts key elements and weights skills.

2. Advanced LangChain Agent: Generates adaptive behavioral, technical, and situational questions, tailoring difficulty based on real-time performance. Provides accuracy feedback.

3. Simulated Interviewer Agent: A single agent simulating a typical interviewer, personalising questions based on the resume and job description.

4. Comprehensive Feedback: Provides automated scores for communication, technical skills, and overall performance. Offers personalised recommendations.

Quality Attributes

1. Adaptability (Primary)

Crucial to tailor content to user needs, experience, and target role. Enables dynamic adjustment of question difficulty, format, and style based on performance, learning style, and skill gaps. Facilitates continuous system learning.

Measurement:

• Difficulty adjusts within 2-3 turns, maintaining 70-85% target accuracy.
• Personalisation algorithms increase engagement by 30% vs. generic sets (measured by session length, repeat usage).
• User performance (score, time per question) improves by 15% post-training.
• Expert SMEs evaluate and confirm that question difficulty properly aligns with user responses.

2. Realism (Primary)

Essential to create immersive simulations with questions and scenarios that mirror real-world interviews reflecting target roles and company cultures with believable interviewer personalities.

Measurement:

• Users rate the experience “realistic” or “very realistic” in 85% of sessions (5-point Likert scale).
• SMEs validate that 95% of content is relevant/representative.
• User feedback deems agent personalities engaging/believable in 80% of sessions.
• Blind studies show users unable to reliably distinguish AI from human interviewers.
• Response times are comparable to a real human interviewer.

3. Relevance (Secondary)

Critical to ensure questions, feedback, and recommendations align with the candidate’s skills, experience, and the job’s requirements.

Measurement:

• 90% of generated questions directly relate to resume skills and job description.
• Feedback provides specific, actionable steps for improvement.
• 80% of users rate questions/feedback as “highly relevant” or “relevant.”
• Relevance scores show generated content align with skills needed in job description.

Evaluation

1. Adaptability Testing

• Performance A/B Test: Compare adaptive vs. static sets. Track engagement (session length, repeat usage) and scores. Success: 30% higher engagement, 15% better scores.
• Personalisation Study: Measure user satisfaction with personalised content, gathering feedback. Success: 85% rate scenarios as “highly relevant.”
• Difficulty Evaluation: SMEs review if question difficulty aligns with user’s performance. Experts assess that user’s good answers lead to harder questions. Data will show users passing initial stages but later failing, proving that questions increased in difficulty.

2. Realism Evaluation

• SME Review Board: Experts (interviewers, HR) evaluate questions, scenarios, agents against a rubric. Success: 95% rated “realistic” and “relevant.”
• Blind User Study: Users compare AI and real interviews, without knowing which is which. Success: Users can’t reliably distinguish them (measure accuracy rates).
• Interview Comparison: Score and compare to real interview; ensure the marks are comparable.

3. Relevance Assessment

• Relevance Score Calculation: Develop automated system scoring relevance of generated questions using candidate’s resume and job description. Success: Average score exceeds 90%.
• Feedback Actionability Analysis: Track user behavior post-feedback to measure changes. Success: Show progress by following personalised recommendations with more improvement than without.
• User Analysis: Collect user feedback and measure relevance.