🎓 Multi-Agent Research Assistant

An advanced AI agent system designed to automate the academic research process. Powered by Google Gemini 2.5, this project uses a multi-agent architecture to search for papers, analyze publication trends, format citations, and persist findings.

🚀 Features

• Multi-Agent Architecture: Specialized agents working in concert: 🕵️ Researcher: Searches ArXiv for relevant papers (supports parallel queries). 📊 Analyst: Analyzes data (e.g., publication year distribution) and generates ASCII charts. 📝 Formatter: Formats raw paper data into proper APA citations. 🧠 Root Agent: Orchestrates the workflow and manages memory.

• Tools & Integrations: ArXiv API: Real-time access to millions of academic papers. Long-Term Memory: Persists findings to a JSON knowledge base. Parallel Execution: Simulates concurrent research tasks for efficiency.

• Deployment Ready: Run locally via Flask API. Deploy to production on Google Cloud Vertex AI Agent Engine. Interactive testing via ADK Web UI.

🏗️ Architecture

The system follows a hierarchical multi-agent pattern:

graph TD
    User[User Query] --> Root["Root Agent (Orchestrator)"]
    Root -->|Delegates Search| Researcher[Researcher Agent]
    Researcher -->|Calls| ArXiv[ArXiv API]
    Root -->|Delegates Analysis| Analyst[Analyst Agent]
    Root -->|Delegates Formatting| Formatter[Formatter Agent]
    Root -->|Saves Data| Memory[(Long-Term Memory)]
    Root -->|Final Report| User

Loading

🛠️ Tech Stack

• LLM: Google Gemini 2.5 Flash Lite
• Framework: Google Agent Development Kit (ADK)
• Language: Python 3.10+
• APIs: ArXiv API
• Deployment: Flask, Vertex AI, Docker

📦 Installation

1. Clone the repository:

   git clone https://github.com/emansarahafi/research-assistant-agent.git
   cd research-assistant-agent

2. Install dependencies:

   pip install google-adk arxiv nest-asyncio google-cloud-aiplatform flask

3. Set up API Keys:

   • You need a Google Cloud Project with Vertex AI enabled.

   • Set your API key as an environment variable:

     export GOOGLE_API_KEY="YOUR_API_KEY"

💻 Usage

Option 1: Jupyter Notebook

Run the research-assistant-agent.ipynb notebook to step through the entire process, from agent definition to execution and deployment.

Option 2: Local Flask API

The project includes a Flask wrapper for local testing.

1. Run the Flask server code (found in Section 7 of the notebook).

2. Send a POST request:

   curl -X POST http://localhost:5003/chat \
     -H "Content-Type: application/json" \
     -d '{"query": "Agentic AI Design Patterns"}'

Option 3: Vertex AI Deployment

1. Generate the agent package (Section 6 of the notebook).

2. Deploy using gcloud:

   gcloud beta ai agents create \
     --display-name="Research Assistant" \
     --project=$PROJECT_ID \
     --location=$LOCATION \
     --agent-package-path="vertex_agent_deployment/"

📂 Project Structure

.
├── research-assistant-agent.ipynb  # Main project notebook
├── vertex_agent_deployment/        # Generated deployment package
│   ├── agent.py                    # Full multi-agent source code
│   ├── requirements.txt            # Python dependencies
│   └── config.json                 # Vertex AI config
├── research-agent/                 # ADK Web UI package
└── knowledge_base.json             # Persistent memory file (generated)

📝 License

This project is open-source and available under the MIT License.

❓ Troubleshooting

• WARNING:google_genai.types: You may see warnings about "non-text parts" in the logs. These are harmless and caused by the agent returning function calls. They are suppressed in the notebook configuration.
• ADK Web UI 500 Error: Ensure you run the !adk web cell before clicking the proxy link.