TalentIQ AI Recruitment Copilot

Overview

TalentIQ is an AI-powered recruitment copilot that revolutionizes the hiring process by intelligently matching candidates to job descriptions. Built as my final year capstone project, it achieved a 40% improvement in candidate-job fit compared to traditional keyword-based systems.

The Problem

Manual resume screening is:

• Time-consuming: Recruiters spend 23 hours screening resumes for a single hire
• Biased: Unconscious bias affects 70% of hiring decisions
• Imprecise: Keyword matching misses 50% of qualified candidates
• Expensive: Bad hires cost companies 30% of annual salary

Technical Solution

1. Resume Parsing Engine

• Extracted structured data from PDF/DOCX resumes using custom NLP pipeline
• Identified skills, experience, education with 95% accuracy using spaCy + custom models
• Normalized entities (job titles, companies, skills) for consistency

2. Semantic Matching with LLMs

• Generated embeddings for resumes and job descriptions using OpenAI Ada-002
• Implemented cosine similarity scoring with weighted skill matching
• Used GPT-3.5-turbo for contextual analysis and explanation generation

3. Streamlit UI + Firebase Backend

• Built interactive Streamlit dashboard for recruiters
• Integrated Firebase Authentication for multi-tenant support
• Real-time updates with Firebase Realtime Database

Key Features

Results

• 40% improvement in candidate-job fit score
• 75% time savings in initial screening phase
• 500+ resumes processed in pilot program
• 4.5/5 user rating from recruiters in feedback survey

Technical Stack

Key Learnings

• Prompt Engineering Matters: Spent 40% of dev time optimizing LLM prompts for consistent, explainable outputs.
• Data Quality > Model Complexity: Clean, structured resume data improved matching more than fancy algorithms.
• User Trust is Hard-Won: Explainability features were critical for recruiter adoption—black box scores weren't enough.
• Ethical AI Design: Bias detection and mitigation required continuous monitoring and feedback loops.