How to Configure Google Gemini API for the Hiring Agent

Set the GEMINI_API_KEY environment variable in your .env file and configure LLM_PROVIDER=gemini to switch the Hiring Agent from Ollama to Google Gemini.

The Hiring Agent by InterviewStreet supports multiple LLM backends, including local Ollama instances and cloud-based Google Gemini models. Configuring Gemini requires setting a single API key and selecting a model name, after which the application automatically routes all inference requests to Google's Generative AI API.

Prerequisites

Before configuring Gemini, ensure your environment has the required dependencies. The repository includes google-generativeai in its requirements.txt for interacting with the Gemini API.

pip install -r requirements.txt

You will also need a valid Google Gemini API key from Google AI Studio.

Environment Configuration

The Hiring Agent uses environment variables to select the LLM provider and authenticate with external APIs. Create a .env file in your project root by copying the provided template:

cp main/.env.example .env

Edit the .env file to include these three critical variables:

LLM_PROVIDER=gemini          # Switches from Ollama to Gemini backend

DEFAULT_MODEL=gemini-2.5-pro # Or gemini-2.5-flash, gemini-1.5-pro, etc.

GEMINI_API_KEY=YOUR_ACTUAL_API_KEY_HERE

The GEMINI_API_KEY is read by prompt.py using python-dotenv's load_dotenv() function, which executes at module import time to populate os.getenv("GEMINI_API_KEY").

How the Configuration Works

Understanding the internal flow helps troubleshoot configuration issues. The Hiring Agent implements a provider pattern that maps model names to specific implementations.

Provider Selection in prompt.py

In main/prompt.py, lines 36-44 and 55-63 define a mapping between model names and the ModelProvider.GEMINI enum. When you set DEFAULT_MODEL to any Gemini model identifier (e.g., gemini-2.5-pro), the system resolves this to use the Gemini backend.

The API key is loaded in lines 12-18, where load_dotenv() processes your .env file, and lines 66-68 store the key in a module-level variable for later use.

GeminiProvider Implementation

The GeminiProvider class in main/models.py (lines 13-21) handles the actual API communication. Its constructor:

1. Receives the API key from prompt.py
2. Calls google.generativeai.configure(api_key=...) to initialize the Google client
3. Stores the configured client for subsequent chat operations

When generating responses, the chat() method (lines 48-66 and 70-78) constructs a GenerativeModel instance using your specified model name. It passes generation parameters—such as temperature and top_p defined in MODEL_PARAMETERS from prompt.py—to the Gemini API and returns standardized response objects compatible with the rest of the Hiring Agent codebase.

Key Validation

The llm_utils.py module validates that GEMINI_API_KEY is not empty before constructing the provider instance. If the environment variable is missing or blank, the code raises an error immediately rather than attempting to call the Gemini API without credentials.

Testing Your Configuration

After updating your .env file, verify the configuration by running the resume evaluation script:

python -m main.evaluate path/to/candidate_resume.json

If configured correctly, the agent will initialize the Gemini provider and process the resume using your selected model. You should see no errors related to missing API keys or provider misconfiguration.

Summary

• Environment Setup: Place GEMINI_API_KEY, LLM_PROVIDER=gemini, and DEFAULT_MODEL in a .env file in your project root.
• Loading Mechanism: main/prompt.py uses python-dotenv to load these variables at runtime, mapping Gemini model names to ModelProvider.GEMINI.
• Provider Implementation: main/models.py contains the GeminiProvider class that configures the Google client via google.generativeai.configure() and handles all chat completions.
• Validation: The system validates key presence in llm_utils.py before execution, preventing empty credential errors.

Frequently Asked Questions

What Gemini models are supported by the Hiring Agent?

The Hiring Agent supports any model identifier recognized by Google's Generative AI API, including gemini-2.5-pro, gemini-2.5-flash, gemini-1.5-pro, and gemini-1.5-flash. These mappings are defined in main/prompt.py and resolve to ModelProvider.GEMINI when selected via the DEFAULT_MODEL environment variable.

Can I configure Gemini without using a .env file?

Yes. You can set the environment variables programmatically before importing the Hiring Agent modules. Since prompt.py reads environment variables at import time, you must set them in Python before importing:

import os
os.environ["GEMINI_API_KEY"] = "your-key-here"
os.environ["LLM_PROVIDER"] = "gemini"
from main.evaluate import evaluate_resume

How do I switch back to Ollama from Gemini?

Change the LLM_PROVIDER environment variable to ollama (or remove it entirely, as Ollama is the default backend) and update DEFAULT_MODEL to an Ollama-supported model name such as llama3.1 or mistral. The provider selection logic in main/prompt.py will automatically route requests to the local Ollama instance instead of the Gemini API.

Why does the Hiring Agent require a Gemini API key instead of using application default credentials?

The GeminiProvider class in main/models.py explicitly initializes the Google client using google.generativeai.configure(api_key=...) rather than Application Default Credentials (ADC). This design choice ensures that users can easily rotate keys via environment variables and supports deployment scenarios where ADC may not be available, such as containerized environments or CI/CD pipelines.

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