Lecture Notes: Building End-to-End Agentic AI Applications

Introduction

• Presenter: Krish Naik
• Focus: Building agentic AI applications using multiple frameworks
• Today's Framework: PhiData (open-source platform)
  • Build, deploy, and monitor agentic systems
  • Create AI agents, multi-modal agents, and agentic workflows
  • Use any LLM (Large Language Model) to create agents

Framework: PhiData

• Features of PhiData:
  • Open-source platform for building agentic systems
  • Supports integration with various LLMs like Grok, HuggingFace, etc.
  • Allows creation of complex workflows
  • Easy code writing for integration
• Integration Examples:
  • Grok, HuggingFace, Ulama, OpenAI
  • Various models: Grok, NVIDIA, Azure, AWS Bedrock

Setting Up the Environment

• Python Environment Setup:
  • Use Conda to create a virtual environment
  • Python version 3.12
• Requirements File (requirements.txt):
  • Libraries to install: PhiData, python-dotenv, YFinance, packaging, DuckDuckGo Search, FastAPI, Uvicorn, Grok

Creating AI Agents

Web Search Agent

• Purpose: Search the web for information
• Components:
  • Model: Grok
  • Tool: DuckDuckGo Search
  • Parameters: Include sources, use markdown

Financial Agent

• Purpose: Analyze financial data
• Components:
  • Model: Grok
  • Tool: YFinance
  • Parameters: Analyze recommendation, company news, technical indicators

Combining Agents

• Multi-Model Agent: Combines web search and financial agents
  • Instructions: Include sources, use tables
  • Parameters: Stream responses

Building the Application

Creating the Project

• File: financial_agent.py
• Library Imports: PhiData, Grok, YFinance, DuckDuckGo
• Environment Variables:
  • Set API keys for Grok and PhiData

Running the Application

• Setup Python environment
• Install dependencies from requirements.txt
• Define and run agents

Error Handling

• Common Errors: OpenAI API key requirement (even if not used)
• Solution: Set OpenAI API key as environment variable

Using the Playground

• Purpose: Visual interface for interacting with agents
• Setup:
  • Connect to localhost endpoint
  • Use FastAPI for creating APIs
• Example Queries:
  • Get analyst recommendations and latest news for stocks
  • Compare stock recommendations between Tesla and NVIDIA

Conclusion

• Outcome: Successfully built and demonstrated financial analysis application
• Next Steps: Explore more complex workflows
• Motivation: Continue to learn and develop more agentic AI solutions

End of Lecture