Overview

Purpose and Scope

TalentSync is an AI-powered hiring intelligence platform that automates resume analysis, candidate evaluation, and job application workflows for both job seekers and recruiters. This document provides a high-level introduction to the system architecture, target users, core capabilities, and technology stack.

For detailed information on specific subsystems:

• Backend service implementation details, see Backend Services
• Frontend application structure, see Frontend Application
• Database schema and data models, see Database & Data Models
• Deployment infrastructure, see Deployment & Infrastructure

System Architecture Overview

TalentSync implements a three-tier architecture consisting of a Next.js frontend, FastAPI backend, and PostgreSQL database, with extensive integration of AI/ML services.

System Components

Frontend Application Stack

Backend Service Stack

Target Users and Use Cases

TalentSync serves two distinct user segments with different feature sets:

Job Seeker Features

Recruiter Features

Core Processing Pipeline

The resume analysis pipeline demonstrates the integration of traditional ML and modern LLM techniques:

Processing Stages

1. Text Extraction (backend/server.py752-793): Handles PDF, DOCX, and TXT formats using PyPDF2 and python-docx libraries
2. Text Cleaning (backend/server.py738-749): Applies spaCy lemmatization and NLTK stopword removal
3. ML Classification (backend/server.py714-735): Uses pre-trained best_model.pkl (25 job categories) with TF-IDF features
4. Field Extraction (backend/server.py923-1029): Regex-based extraction of name, email, skills, work experience, projects
5. LLM Enhancement (backend/server.py373-462): Google Gemini 2.0 Flash generates structured analysis via ComprehensiveAnalysisData schema
6. Storage (Prisma ORM): Persists both raw text and structured analysis in PostgreSQL

Processing Time: Typically 4-8 seconds for complete analysis including LLM calls.

AI/ML Technology Stack

TalentSync employs a hybrid approach combining traditional machine learning with large language models:

LLM Configuration

The system uses Google's Gemini 2.0 Flash model with the following configuration:

# From backend/server.py:76-80
llm = ChatGoogleGenerativeAI(
    model="gemini-2.0-flash",
    google_api_key=google_api_key,
    temperature=0.1,
)

Temperature: Set to 0.1 for deterministic, factual outputs suitable for structured data extraction.

Data Models

Pydantic Schemas

The backend uses Pydantic for request/response validation and data structuring:

Database Schema

The PostgreSQL database uses Prisma ORM for type-safe access. Key tables include:

• User: Authentication and profile data
• Resume: Uploaded resume files and metadata
• Analysis: Structured analysis results linked to resumes
• Session: NextAuth session management
• VerificationToken: Email verification tokens

For complete schema details, see Database & Data Models.

Deployment Architecture

TalentSync is deployed as a multi-container Docker Compose application:

Container Specifications

CI/CD Pipeline

GitHub Actions automatically deploys on push to main branch:

1. SSH into VPS
2. Pull latest code
3. Rebuild Docker images
4. Restart containers with zero downtime

For detailed deployment configuration, see Deployment & Infrastructure.

Key Differentiators

TalentSync distinguishes itself through several technical capabilities:

System Metrics

Based on platform usage as displayed on landing pages:

Market Context (from about page):

• AI in HR market: $6.05B (2024) → $14.08B (2029), 19.1% CAGR
• Resume parsing market: $20.19B (2024) → $43.20B (2029), 114% growth

Next Steps

For detailed information on specific subsystems:

• Backend Implementation: See Backend Services for API endpoints, ML services, and agent architecture
• Frontend Features: See Frontend Application for page structure, authentication, and UI components
• Data Layer: See Database & Data Models for schema design and relationships
• Operations: See Deployment & Infrastructure for Docker setup, CI/CD, and monitoring