AI Memory
Short-term Memory
- Current conversation
Long-term Memory
- Stored interactions
Context Window
- Max tokens processed
- Uses FIFO (First In, First Out)
Conversation ID
- Enables memory across sessions
RAG – Deep Dive
Steps:
- Retrieval
- Augmentation
- Generation
Knowledge Files vs Tools
| Knowledge Files | Tools |
|---|---|
| Full context | Targeted retrieval |
| Static data | Dynamic access |
Direct File Processing
- Supports:
- Images
- Text
MCP Architecture
Core Elements
- Tools → Actions
- Resources → Data
- Prompts → Templates
MCP Communication
Uses:
- JSON-RPC
- Requests / Responses / Notifications
Security
- API Keys
- OAuth 2.0
Real-World AI Agent Examples
1. Customer Invoice Agent
- Detects duplicate charges
- Issues refunds
2. Meeting Assistant
- Reads calendar
- Calculates travel time
- Sends alerts
3. Sales Assistant
- Manages CRM
- Generates reports
4. Security Agent
- Monitors roles
- Handles onboarding
Why Agentic Automation?
- Handles complex tasks
- Works with unstructured data
- Makes real-time decisions
- Uses human-like reasoning
Final Thought
AI is evolving from:
- Tools → Assistants → Autonomous Agents
Understanding LLMs and systems like MCP and RAG gives you the ability to:
- Build real systems
- Automate workflows
- Create AI-driven businesses