Expected Input Format for evaluator.py in InterviewStreet's Hiring Agent

The evaluator.py module expects a single plain-text string containing the full content of a candidate's resume, passed to the evaluate_resume() method as the resume_text argument.

The evaluator.py file in the interviewstreet/hiring-agent repository defines the ResumeEvaluator class, which powers AI-driven resume screening. Understanding the expected input format for evaluator.py is essential for integrating this component into your hiring pipeline, as the system processes raw textual data rather than structured documents or binary files.

The Core Input: Plain Text Resume

The primary entry point for resume evaluation is the evaluate_resume method in evaluator.py. This method accepts exactly one required positional argument:

• resume_text (str): The complete textual content of the candidate's resume.

When you call evaluate_resume(resume_text), the implementation immediately stores the raw string in self._last_resume_text for internal reference. This string is then injected into the prompt templates without modification, which means the system expects clean, extractable text—such as content you would copy-paste from a PDF or Word document—rather than file paths, base64-encoded data, or JSON objects.

No additional structure, markup, or metadata wrappers are required. The LLM handles the unstructured text and extracts relevant evaluation criteria based on the rendered prompts.

Optional Configuration Parameters

While the resume content itself must be a plain string, the ResumeEvaluator class accepts optional initialization parameters that control the underlying LLM behavior. These are defined in the class constructor and referenced from prompt.py:

Parameter	Type	Default	Description
model_name	str	DEFAULT_MODEL (imported from prompt.py)	Specifies the LLM identifier (e.g., "gpt-4o", "gemini-1.5-flash") used for evaluation.
model_params	dict	MODEL_PARAMETERS[model_name] or {"temperature": 0.5, "top_p": 0.9}	Dictionary of inference settings including temperature and sampling parameters.

These parameters affect how the resume text is processed by the model but do not change the input format of the resume itself.

How the Input is Processed

Once you pass the resume text to evaluate_resume, the method executes a structured pipeline defined across several modules:

1. Storage: The raw resume_text is cached in self._last_resume_text.

2. Prompt Rendering: The method calls the template manager from prompts/template_manager.py to render the resume_evaluation_criteria template, substituting the resume text into the prompt.

3. System Context: It retrieves the resume_evaluation_system_message template to establish the LLM's evaluation persona.

4. LLM Invocation: The code in evaluator.py constructs a chat payload containing the system message, the user prompt (with embedded resume text), and the model_params. This is sent via initialize_llm_provider from llm_utils.py.

5. Response Parsing: The method expects a JSON-encoded string representing an EvaluationData object (defined in models.py), which it parses and returns as a structured Pydantic model.

Complete Working Example

Below is a runnable example demonstrating the expected input format and class instantiation:

from evaluator import ResumeEvaluator

# 1️⃣ Create the evaluator (optional model customization)

evaluator = ResumeEvaluator(
    model_name="gpt-4o-mini",        # any supported model name

    model_params={"temperature": 0.3, "top_p": 0.95}
)

# 2️⃣ Prepare the resume text (plain string)

resume_text = """
John Doe
Software Engineer
Experience:
- Developed microservices in Python and Go
- Led a team of 5 engineers
Education:
- B.Sc. Computer Science, XYZ University
Skills: Docker, Kubernetes, AWS, REST APIs
"""

# 3️⃣ Run the evaluation

evaluation = evaluator.evaluate_resume(resume_text)

# 4️⃣ Access the structured result

print(evaluation.json())

Summary

• Input Type: evaluator.py requires a single Python string (resume_text) containing the full plain-text content of the resume.
• No Preprocessing Needed: File paths, binary PDFs, or Word documents must be converted to text before passing to evaluate_resume().
• Configuration: Initialize ResumeEvaluator with optional model_name and model_params to customize LLM behavior without changing input format.
• Output: The method returns a parsed EvaluationData Pydantic model from models.py, derived from JSON output by the LLM.
• Key Files: evaluator.py (core logic), prompt.py (default models), prompts/template_manager.py (template rendering), models.py (output schema), and llm_utils.py (provider initialization).

Frequently Asked Questions

Can evaluator.py process PDF or Word documents directly?

No. The evaluate_resume method in evaluator.py accepts only Python strings. You must extract text from PDF, DOCX, or other binary formats using external libraries (such as PyPDF2 or python-docx) before passing the content as the resume_text argument.

What structure should the resume text follow?

The input should be raw, unstructured plain text without special markup requirements. The ResumeEvaluator class sends this text to the LLM within predefined prompt templates from prompts/template_manager.py, so standard resume formatting with line breaks and bullet points works optimally.

Is there a maximum length limit for the resume text?

The effective limit depends on the model_name you specify during initialization and its associated context window constraints defined in prompt.py. If the resume exceeds the model's token limit, the LLM provider call will fail, so you should verify the context length for your chosen model (e.g., GPT-4o, Gemini-1.5-Flash).

What does the evaluate_resume method return?

According to the implementation in evaluator.py, the method returns an EvaluationData object as defined in models.py. This Pydantic model parses the JSON response from the LLM, providing structured access to evaluation scores, criteria matches, and candidate assessments.

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