Generify - Drug Cost Comparison Agent

A tool-using AI agent that helps users find equivalent drugs and rank them by Medicare Part D average spend per dosage unit. Built with Google ADK and Gemini.

🎯 Overview

Generify uses the Orange Book (FDA) to identify therapeutic equivalent drugs and Medicare Part D data to rank them by cost-effectiveness. The agent helps users find the most cost-effective generic alternatives to brand-name medications.

📋 Submission Checklist

• All code in src/ runs without errors
• ARCHITECTURE.md contains a clear diagram sketch and explanation
• EXPLANATION.md covers planning, tool use, memory, and limitations
• DEMO.md links to a 3–5 min video with timestamped highlights (https://youtu.be/BxoGwARAWCY)

🚀 Getting Started

Prerequisites

• Python 3.11+
• Google Gemini API key (Get one here)
• Source data files (see Data Setup below)
• Google ADK with eval

Installation

0. Install Google ADK pip install 'google-adk[eval]'

1. Clone the repository

   git clone https://github.com/Akhil3171/Generify.git
   cd Generify

2. Install dependencies

   pip install -r requirements.txt

3. Configure API key

   ADK automatically reads GOOGLE_API_KEY from environment variables. You can either:

   • Use a global environment variable (recommended)
   • Or create a .env file in the project root:

     GOOGLE_API_KEY=your_api_key_here

   Get your API key from: https://aistudio.google.com/app/apikey

4. Build databases (if not already present)

   Download the required source data files:

   • Orange Book data: products.txt → place in Data/ folder
   • Medicare Part D data: DSD_PTD_RY25_P04_V10_DY23_BGM.csv → place in Data/ folder

   Then run:

   python build_db.py

   This creates:

   • Data/products.db (Orange Book database)
   • Data/medicare.db (Medicare Part D database)

5. Run the ADK web server

   adk web .

   The server will start and display a URL (typically http://127.0.0.1:8080). Open it in your browser.

   Alternative commands:

   # Custom port
   adk web . --port 8081
   
   # Verbose logging
   adk web . -v
   
   # Auto-reload (development)
   adk web . --reload

6. OPTIONAL - Run adk eval test cases e.g. on separate terminal or new command line a. Run tests and get simple results (PASS/FAIL metrics)

adk eval drug_cost_agent evaluation\test_cases.json --config_file_path evaluation\eval_config.json

b. Run tests and get detailed results

adk eval drug_cost_agent evaluation\test_cases.json --config_file_path evaluation\eval_config.json --print_detailed_results

Project Structure

Generify/
├── drug_cost_agent/          # ADK Agent (Web Interface)
│   ├── __init__.py
│   └── agent.py              # Agent definition using Gemini
├── src/                      # Core Code (Required Architecture)
│   ├── planner.py            # Task decomposition (Gemini API)
│   ├── executor.py           # Tool execution (Gemini API)
│   ├── memory.py             # Session storage
│   ├── agent_core.py         # Workflow orchestrator
│   ├── paths.py              # Database paths
│   ├── tools/                # Tools directory
│   │   ├── tools_ob.py       # Orange Book tools
│   │   ├── tools_medicare.py # Medicare Part D tools
│   │   └── memory_tools.py  # Memory tools (ADK)
│   └── plugins/              # ADK plugins
├── Data/                     # Databases & Source Data
│   ├── products.db           # Orange Book (built)
│   ├── medicare.db           # Medicare Part D (built)
│   └── sessions.json         # Session memory (auto-created)
├── build_db.py               # Database builder
├── requirements.txt          # Dependencies
└── .env                      # API key (optional, uses env vars)

See ARCHITECTURE.md and EXPLANATION.md for detailed technical explanations.

Usage

Once the ADK web server is running, interact with the agent by asking questions like:

• "Find equivalent drugs for Lipitor 20mg"
• "What are the cheapest alternatives to Advil?"
• "Compare costs for metformin 500mg tablets"
• "I need a cheaper option for Prozac"

Troubleshooting

Agent not found:

• Ensure you're in the project root directory
• Run adk web . (note the dot)
• Verify drug_cost_agent/ folder exists

API key errors:

• Check that GOOGLE_API_KEY is set in environment variables
• Test: python -c "import os; print(os.getenv('GOOGLE_API_KEY'))"

Database errors:

• Ensure databases exist: ls Data/*.db
• Rebuild if needed: python build_db.py

Import errors:

• Test imports: python -c "from drug_cost_agent import root_agent"
• Verify dependencies: pip install -r requirements.txt

Port already in use:

• Use different port: adk web . --port 8081
• Or kill process using port 8080

Visual representation of the project directory structure showing the organization of ADK agent, source code, tools, and data files.