Resume Matching Engine

Redrob AI Campus Hackathon — Individual Competition

---

Problem Statement

Built a Resume Matching Engine that:

• Normalizes noisy resume skill data from 10 Indian university students
• Computes TF-IDF vectors for resumes
• Builds binary vectors for 3 Job Descriptions from Korean tech companies
• Calculates cosine similarity between resumes and JDs
• Outputs the Top 3 matching candidates per JD

---

How It Works

Step 1 — Skill Normalization

• Split raw skills on commas
• Convert to lowercase
• Match multi-word phrases before single tokens (sorted by length descending)
• Apply SKILL_ALIASES mapping exactly as provided
• Discard tokens not present in the alias map

Step 2 — Deduplication

• Each canonical skill appears only once per resume

Step 3 — Vocabulary Construction

• Built from normalized + deduplicated resume skills only
• Sorted alphabetically (48 unique terms)

Step 4 — TF-IDF Vectors (Resumes only)

TF  = 1 / N               (N = total unique skills in resume)
IDF = ln(10 / df)         (natural log, no smoothing)
TF-IDF = TF × IDF

Step 5 — JD Binary Vectors

• 1 if skill present in JD, 0 if not
• Built over same shared vocabulary

Step 6 — Cosine Similarity & Ranking

Cosine(A, B) = (A · B) / (|A| × |B|)

• A = Resume TF-IDF vector
• B = JD binary vector
• Top 3 ranked per JD, ties broken alphabetically

---

Results

JD-1 — Kakao (ML Engineer)
Sneha Patel(0.57), Karan Mehta(0.53), Arjun Sharma(0.40)

JD-2 — Naver (Backend Engineer)
Rahul Gupta(0.81), Ananya Krishnan(0.28), Deepika Rao(0.19)

JD-3 — Line (Frontend Engineer)
Aditya Kumar(0.67), Priya Nair(0.58), Ananya Krishnan(0.35)

---

Tech Stack

Item	Detail
Language	Python
Libraries	Standard library only (math)
External Libraries	None (not allowed)
AI Tool Used	Redrob AI

---

File Structure

├── resume_matcher.py   # Main solution file
└── README.md           # This file

---

How to Run

Google Colab:

1. Open colab.research.google.com
2. Create a new notebook
3. Paste resume_matcher.py contents into a cell
4. Press Shift + Enter to run

VS Code / Terminal:

python resume_matcher.py

No installations needed — uses Python standard library only.

---

Rules Followed

• ✅ Only standard library used (math) — no numpy, pandas, sklearn
• ✅ SKILL_ALIASES used exactly as provided, not modified
• ✅ Multi-word phrases matched before single tokens
• ✅ Vocabulary built from resume skills only
• ✅ TF-IDF computed for resumes only, not JDs
• ✅ IDF = ln(10/df), natural log, no smoothing
• ✅ Cosine similarity uses Euclidean norm of resume vector
• ✅ Scores rounded to 2 decimal places
• ✅ Ties broken alphabetically by candidate name

---

Redrob AI Campus Hackathon · Powered by McKinley Rice