Agentic AI Explained | Lecture 7 Part 1 – How AI Agents Work (LLM, Memory, Tools) скачать в хорошем качестве

Agentic AI Explained | Lecture 7 Part 1 – How AI Agents Work (LLM, Memory, Tools)

Lecture 7 – AI Agents Deep Dive (Part 1) 1. What is an AI Agent? An AI Agent is not just a chatbot or a single AI call. It is a decision-making system that can: Understand inputs Reason over information Use tools Store and recall memory Take actions Produce structured outputs In simple terms: An AI agent is an intelligent system that can think, decide, remember, and act to achieve a goal. 2. Core Components of an AI Agent Every production-grade AI agent is built from multiple components. Removing any one of them limits the agent’s capability. 3. LLM (Large Language Model) Role of the LLM The LLM is the brain of the AI agent. It is responsible for: Understanding natural language Reasoning and decision-making Generating responses Interpreting context Examples: OpenAI models Gemini models Important concept: The LLM does not know your business by default. It only knows general knowledge unless connected to memory or tools. 4. Tools What are Tools? Tools allow the AI agent to interact with the outside world. Examples of tools: APIs Databases Web scrapers Search engines Internal workflows (n8n nodes) Without tools: AI can only talk With tools: AI can act Examples: Fetch customer data Scrape a website Send messages Store information Trigger workflows 5. Memory in AI Agents Why Memory is Needed Without memory: The agent forgets everything after each interaction No personalization No learning from past conversations Memory allows agents to: Remember users Track conversations Maintain long-term context Improve decisions over time 6. Types of Memory 1. Short-Term Memory Context of the current conversation Stored temporarily Lost after the session ends 2. Long-Term Memory Stored permanently Used across multiple sessions Enables personalization and continuity 7. Supabase as Agent Memory Why Supabase? Supabase provides: PostgreSQL database Authentication Storage Real-time updates It is commonly used as long-term memory for AI agents. Use Cases: Store user profiles Save conversation history Track agent decisions Persist structured data Supabase enables agents to: Recall previous interactions Maintain state Act intelligently over time 8. PostgreSQL (Postgres) in Agentic AI Role of PostgreSQL PostgreSQL is a relational database used to store: Structured memory Logs User data Agent states Why Postgres is important: Reliable Scalable Widely supported Works perfectly with Supabase In Agentic AI: Postgres acts as the backbone for persistent memory. 9. RAG-Based Agents What is a RAG Agent? A RAG (Retrieval-Augmented Generation) Agent combines: Memory External knowledge AI reasoning Instead of guessing answers, the agent: Retrieves relevant data Passes it to the LLM Generates accurate responses Why RAG is Critical Prevents hallucinations Keeps responses factual Separates knowledge from the model Allows easy updates RAG agents are commonly used for: Customer support Knowledge bases Internal company assistants 10. Output Parser What is an Output Parser? An output parser ensures that AI responses are: Structured Predictable Machine-readable Instead of free text, outputs can be: JSON Key-value pairs Clean formatted data Why Output Parsers Matter Prevents messy outputs Enables automation Makes AI responses usable in workflows Example: Instead of: “Sure, here is the answer…” The agent returns: { "intent": "order_status", "confidence": 0.92, "action": "fetch_order" } 11. Full AI Agent Architecture (Conceptual Flow) User Input LLM interprets request Agent checks memory (Supabase/Postgres) Agent retrieves knowledge (RAG if needed) Agent calls tools Output parser structures response Final action or response is produced This architecture allows agents to behave like real digital workers, not chatbots. 12. Key Takeaways from Lecture 7 – Part 1 AI agents are multi-component systems LLMs provide reasoning, not business logic Tools enable real-world actions Memory makes agents intelligent over time Supabase + PostgreSQL provide long-term memory RAG improves accuracy and reliability Output parsers make AI usable in automation