How Hiring Agent Parses Different Sections of a Resume

Hiring Agent extracts resume sections by converting PDF documents to raw text, then processing that text through section-specific LLM prompts defined in Jinja templates, and finally aggregating the structured responses into a Pydantic JSONResume model.

The interviewstreet/hiring-agent repository implements a hybrid parsing pipeline that transforms unstructured candidate documents into structured data. This technical approach combines PDF text extraction with AI-driven content classification to reliably identify and separate distinct resume sections like work experience, education, and skills.

PDF Text Extraction Foundation

The parsing process begins in main/pdf.py with the PDFHandler class, which handles the initial document ingestion. The extract_text method (around line 54) reads the uploaded PDF file, converts each page to raw text, and concatenates the results into a single continuous string.

This raw text extraction serves as the contextual foundation for all subsequent section parsing. The system does not attempt to split sections during this phase; instead, it preserves the full document text to maintain contextual relationships between different parts of the resume.

Section-Specific LLM Prompting Strategy

Rather than using rule-based parsing, Hiring Agent employs a template-driven LLM strategy defined in main/prompts/template_manager.py. This module maintains a dictionary of Jinja templates (such as basics.jinja, work.jinja, education.jinja, skills.jinja, projects.jinja, and awards.jinja) that instruct the LLM how to extract specific sections.

Each template defines the expected JSON schema for its respective section while embedding the complete raw resume text as context. The _call_llm_for_section method in main/pdf.py (around line 145) orchestrates these calls by:

1. Rendering the appropriate Jinja template with the resume text
2. Sending the formatted prompt to the LLM
3. Parsing the JSON response into the corresponding Pydantic section model

Individual Section Extraction Methods

The PDFHandler class exposes specific extraction methods for each standard resume section:

• extract_basics_section – Parses candidate contact information and personal details
• extract_work_section – Identifies employment history and job responsibilities
• extract_education_section – Extracts degrees, institutions, and academic dates
• extract_skills_section – Catalogs technical and soft skills
• extract_projects_section – Captures portfolio and personal project details
• extract_awards_section – Recognizes certifications and honors

Each method follows the same pattern: render the section-specific prompt, call the LLM via _call_llm_for_section, and return a validated Pydantic model representing that particular section's data structure.

Structured Data Aggregation

Once individual sections are extracted, the system aggregates them into a unified JSONResume model defined in main/models.py (around line 201). This Pydantic model mirrors the official JSON Resume schema and includes typed attributes for basics, work, education, skills, projects, awards, and other standard resume components.

The JSONResume model enforces type safety and validation across all extracted sections, ensuring that the final output conforms to a predictable schema regardless of the input document's original formatting or layout variations.

Converting to Human-Readable Formats

After_structuring the data_, Hiring Agent can transform the JSONResume object into alternative formats for evaluation or export. The convert_json_resume_to_text function in main/transform.py (around line 744) walks through every populated field and produces a human-readable text block.

This text representation feeds into the evaluation pipeline in main/score.py, where it serves as context for assessment LLMs, or exports to CSV format for spreadsheet analysis.

Complete Implementation Example

The following example demonstrates the complete parsing workflow from PDF to structured data:

from main.pdf import PDFHandler
from main.models import JSONResume
from main.transform import convert_json_resume_to_text

# Initialize the PDF handler

pdf_handler = PDFHandler()

# Step 1: Extract raw text from PDF

raw_text = pdf_handler.extract_text("candidate_resume.pdf")

# Step 2: Parse individual sections using LLM-powered extraction

basics = pdf_handler.extract_basics_section(raw_text)
work = pdf_handler.extract_work_section(raw_text)
education = pdf_handler.extract_education_section(raw_text)
skills = pdf_handler.extract_skills_section(raw_text)
projects = pdf_handler.extract_projects_section(raw_text)
awards = pdf_handler.extract_awards_section(raw_text)

# Step 3: Aggregate into unified JSONResume model

resume = JSONResume(
    basics=basics,
    work=work,
    education=education,
    skills=skills,
    projects=projects,
    awards=awards,
)

# Step 4: Convert to text for evaluation or export

resume_text = convert_json_resume_to_text(resume)
print(resume_text)

Summary

• PDF Processing: The PDFHandler class in main/pdf.py converts documents to raw text using the extract_text method.
• LLM-Driven Extraction: Section-specific Jinja templates in main/prompts/template_manager.py guide the LLM to extract structured data via _call_llm_for_section.
• Typed Methods: Dedicated methods like extract_work_section and extract_education_section handle each resume component individually.
• Schema Validation: The JSONResume Pydantic model in main/models.py aggregates all sections into a standardized format.
• Output Flexibility: convert_json_resume_to_text in main/transform.py enables conversion to human-readable formats for scoring or CSV export.

Frequently Asked Questions

What file format does Hiring Agent require for resumes?

Hiring Agent processes PDF documents as its primary input format. The extract_text method in main/pdf.py specifically handles PDF parsing to produce the raw text required for LLM processing.

How does Hiring Agent handle different resume layouts and formats?

The system uses context-aware LLM prompts rather than rigid template matching. By providing the complete resume text to section-specific Jinja templates, the LLM can identify relevant information regardless of formatting variations, column layouts, or section ordering.

Can Hiring Agent extract sections from image-based PDFs or scanned documents?

The analysis indicates text-based extraction via standard PDF parsing. For image-based or scanned PDFs, the repository would require OCR (Optical Character Recognition) preprocessing before the extract_text method could process the content, though this specific capability isn't detailed in the core parsing logic.

What data schema does the parsed resume follow?

Hiring Agent implements the JSON Resume standard through the JSONResume Pydantic model in main/models.py. This schema includes standardized fields for basics, work experience, education, skills, projects, awards, and other common resume sections, ensuring interoperability with other HR tech systems.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}