This project is an AI research assistant that implements agentic workflows using LangChain. The system autonomously decides which tools to use - web search, Wikipedia, or file saving - based on the research query.

Installation

1. Clone the repository

git clone https://github.com/aconstas/ai_agent.git
cd ai_agent

2. Set Up Virtual Environment

python -m venv venv

#Activate virtual environment
# Windows: venv\Scripts\activate
# macOS/Linux: venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

4. Environment Configuration

Create a .env file in the root directory and add your API keys:

ANTHROPIC_API_KEY=your_anthropic_api_key_here
OPENAI_API_KEY=your_openai_api_key_here

Usage

1. Execute the following command:

python main.py

2. Then type your research query in the terminal and press Enter.

3. You will see a file named "research_output.txt" in the repository's root folder. (There is an example "research_output.txt" in the repository.)

Pydantic

Pydantic is a data validation library that ensures data matches a specific structure and format -- like a contract.

In this project, we inherit from Pydantic's BaseModel class:

    class ResearchResponse(BaseModel):
        topic: str
        summary: str
        sources: list[str]
        tools_used: list[str]

This data validation is important to include in the project because AI models are currently notorious for producing unpredictable output formats.

If we were not using Pydantic, the underlying model may produce something like:

"Here's what I found about climate change. The topic is climate change and I used search and wikipedia tools. Here are some sources: nasa.gov, epa.gov..."

With Pydantic, it forces the AI to return structured data like so:

{
    "topic": "climate change",
    "summary": "Climate change refers to long-term shifts...",
    "sources": ["nasa.gov", "epa.gov", "ipcc.ch"],
    "tools_used": ["search", "wiki_tool"]
}

This line creates instructions that get sent to the AI telling it exactly how to format its response:

parser = PydanticOutputParser(pydantic_object=ResearchResponse)

PydanticOutputParser is a LangChain class that acts as a bridge between AI models and structured data. It works like a translator that:

1. Tells the AI how to format its response.
2. Validates the AI's response matches that format.
3. Converts the raw text into a usable Python object.