AI for Engineers

Core Components of an LLM Application

1. Adding RAG (Retrieval-Augmented Generation): This is a key technique for connecting an LLM to your specific data. The process includes:
   • Loading & Ingestion: Using connectors from LlamaHub to import data from various sources (PDFs, APIs, databases).
   • Indexing & Embedding: Structuring the data for efficient and relevant retrieval.
   • Storing: Saving the indexed data, often in a specialized vector store.
   • Querying: Retrieving the most relevant information to answer questions accurately.
2. Building Agents: These are described as LLM-powered "knowledge workers" that can interact with the world using a set of tools. Key aspects include:
   • Creating Agents: Building single agents that can perform tasks.
   • Using Tools: Integrating pre-built tools from the LlamaHub registry.
   • Advanced Features: Maintaining state, streaming output for user feedback, and incorporating a "human in the loop" for verification.
3. Building Workflows: This is a lower-level, event-driven framework for creating advanced and complex agentic systems.
   • Multi-Agent Systems: Using AgentWorkflow to enable multiple agents to collaborate.
   • Control Flow: Implementing logic like looping, branching, and concurrent execution.
   • State Management: Creating stateful workflows that remember information across steps.

Resources

1. Machine Learning Systems

[0]. https://github.com/Nyandwi/machine_learning_complete/blob/main/010_mlops/1_mlops_guide.md [1]. https://dcai.csail.mit.edu/ [2]. DataCentricAI : https://youtube.com/playlist?list=PLnSYPjg2dHQKdig0vVbN-ZnEU0yNJ1mo5&si=OyEQkZEEO3AqKw0j [3]. https://madewithml.com/ [4]. https://stanford-cs329s.github.io/ [5]. https://fullstackdeeplearning.com/