Resume Parser & Evaluator

[]

A production-grade PDF resume parser that extracts structured candidate data, scores resumes across multiple dimensions, and ranks applicants for recruitment pipelines.

---

🚀 Deployment

This system is deployed as a split production architecture.

Backend (FastAPI Resume Parser)

• Hosted on Hugging Face Spaces (Docker)
• Handles PDF ingestion, OCR extraction, parsing, and scoring
• Exposes REST API endpoints for integration

Live Backend: https://huggingface.co/spaces/abzyvantae/Parse_Your_Resume_Backend

---

Frontend (Candidate Dashboard)

• Hosted on Vercel
• Provides resume upload interface and ranking dashboard
• Consumes backend API for processing and data retrieval

Live Frontend: https://web-sigma-flame-90.vercel.app/

---

System Architecture

Frontend (Vercel) │ ▼ FastAPI Backend (Hugging Face Space) │ ├── OCR Engine (Tesseract / fallback) ├── Parsing Layer (regex + rule engine) ├── LLM Enrichment (Groq API) │ ▼ Neon PostgreSQL Database │ ▼ Ranked Candidate Store + Export Layer

---

Deployment Notes

• Backend runs in a Docker container on Hugging Face Spaces
• Cold starts may occur after inactivity
• OCR requires Tesseract installed in runtime environment
• Frontend is stateless and depends on backend availability
• Database operations handled via Neon PostgreSQL

---

Environment Breakdown

Layer	Platform
API Backend	Hugging Face Spaces
Frontend UI	Vercel
Database	Neon PostgreSQL
LLM Provider	Groq API

---

Features

Feature	Description
Multi-engine PDF extraction	PyMuPDF → pdfplumber → OCR fallback chain
Hybrid parsing	Rule-based + LLM semantic analysis
Role-aware scoring	Different weighting for design, marketing, general roles
Behance integration	Portfolio enrichment for creative candidates
Rate-limit resilient	Retry logic with exponential backoff
Unicode-safe	Supports multilingual resumes
OCR fallback	Handles scanned documents

---

Quick Start

Installation

pip install pymupdf requests beautifulsoup4 pytesseract pillow pdf2image groq

---

Basic Usage

python resume_parser.py --input resume.pdf
python resume_parser.py --input ./resumes/

---

API Service

uvicorn api_service:app --host 0.0.0.0 --port 8000

Endpoints:

• GET /healthz
• POST /parse
• POST /parse-batch
• GET /candidates
• PATCH /candidates/{id}/shortlist
• DELETE /candidates/{id}

---

Output Schema

Each resume returns structured JSON:

{
  "name": "John Doe",
  "email": "john@example.com",
  "skills": ["python", "sql", "docker"],
  "ranking_score": 72.5,
  "experience": 6.0,
  "seniority_level": "senior"
}

---

Architecture Flow

PDF Input ↓ Text Extraction (PyMuPDF → pdfplumber → OCR) ↓ Cleaning & Normalization ↓ Regex Feature Extraction ↓ LLM + Rule-Based Parsing ↓ Scoring Engine ↓ Neon Database Storage ↓ API Response

---

Scoring System

Experience, skills, seniority, and quality are combined into a weighted ranking score.

• Experience: nonlinear scaling
• Skills: count-based saturation curve
• Seniority: role inference mapping
• Quality: capped contribution to prevent bias

---

Constraints

• OCR fallback used only when text extraction fails
• Ranking normalization prevents score clustering
• Unicode-safe parsing for international resumes
• Rate limiting handled via retry backoff

---

Testing

120+ test cases covering:

• parsing accuracy
• OCR fallback
• scoring stability
• edge case handling
• API reliability

---

License

ALL RIGHTS RESERVED. Property of Digitalis Global. Unauthorized reproduction, distribution, or modification is strictly prohibited.

Developer: Abeera Tahir