Courses

• Fundamentals of Agents (by Hugging Face). lib: smolagents, LangChain, LlamaIndex.

Why bother?

• Why Agentic RAG over traditional RAG?
  • Traditional RAG:
    • constrained by static workflows, lack the adaptability required for multi-step reasoning and complex task management.
    • Use LLMs for straightforward QA.
  • Agentic RAG: represents an advanced approach that enhances traditional RAG systems by integrating intelligent agents capable of autonomous decision-making and multi-step reasoning.
  • employs specialized agents to navigate complex queries requiring detailed analysis and planning.
  • These agents function similarly to expert researchers, adeptly retrieving relevant information from multiple sources, comparing data, and synthesizing comprehensive responses.

Applications

• Research assistance, customer support, knowledge management.

Components

• Tool/Retrieval Layer:
  • Functionalities: Web Search, APIs, Op. Data, SaaS, Vector DB, Knowledge, Bus. Logic, User Int.
  • Tools: SingleStore, Fast API
• Action / Orchestration Layer:
  • Functionalities: Task Management, Persistent Memory, Automation Scripts, Event Logging.
  • Tools: crewai, LangGraph. AG.
• Reasoning Layer.
  • Functionalities: LLM, Contextual Analysis, Decision Trees, NLU.
  • Tools: GPT, spaCy, scikit-learn
• Feedback / Learning Layer
  • Functionalities: User Feedback Loop, Model Training, Performance Metrics, Continuous Improvement
  • Tools: LangSmith
• Security / Compliance Layer.
  • Functionalities: Data Encryption, Access Control, Compliance Monitoring, Audit Trails.
  • Tools:
• Integration Frameworks:
  • LangChain: building LLM applications.
  • LlamaIndex: data connection and retrieval
  • CrewAI: agent orchestration
  • Semantic Kernel: Microsoft’s orchestration framework
  • AutoGen: multi-agent conversations
  • DSPy: LLM prompt engineering and chaining
  • Haystack: building search and QA pipelines.

Agentic Startups

• The Rise of the Agentic Workforce.

Multi-Agent Systems

• Multi Agent Debate - Rakesh Gohel.
• Supercharge your RAG with Multi-Agent Self-RAG. Based on Self-RAG strategy from LangGraph.

Resources

• RPA vs Agentic Process Automation (APA) vs AI Agents. Fb post.
• 5 most popular Agentic design patterns.
• LangChain Alternatives You Can Use to Build AI and Agentic Workflows