Part 2: AI Agent Code Walkthrough

Episode 2 - AI Agent Code Walkthrough In Episode 1, we covered the theory behind AI agents and autonomy. Now we open up the actual codebase and walk through every layer of a fully working AI Personal Assistant — line by line. By the end of this video, you'll understand exactly how to build an AI agent that reads your emails, manages your calendar, searches your Google Drive, and remembers your preferences — all from a single chat interface. 🔥 What you'll learn: 00:00 - Live Demo 02:00 - Welcome & Intro 05:00 - Architecture Overview (4 Layers) 08:30 - Project Structure Walkthrough 10:00 - The Agent Core: Agentic Loop Pattern 16:30 - Tools: Teaching the Agent New Skills 21:00 - The 3-Tier Memory System 26:30 - System Prompt Engineering 29:40 - Production Backend with Firebase Cloud Functions 34:00 - Frontend Chat Experience 35:30 - End-to-End Flow Walkthrough 39:00 - 5 Essential Patterns & Pro Tips 🧠 5 Key Patterns Covered: 1. The Agentic Loop 2. Dynamic Tool Registration 3. Multi-Tier Memory — short-term, long-term, episodic 4. Dynamic System Prompts — context-aware prompt building 5. Transparency — show users what tools the agent used 🛠️ Tech Stack: - TypeScript + Next.js 15 + Tailwind CSS - Anthropic Claude SDK (Claude AI) - Firebase Cloud Functions + Firestore - Google APIs (Gmail, Calendar, Drive) via OAuth 2.0 📦 Resources: - GitHub Repo: https://github.com/raffimd2/personal-assistant - Episode 1 — What Are AI Agents?: https://www.youtube.com/watch?v=G1k7a2jQgLU - Gen AI Mentor - https://www.genai-mentor.ai/ 👥 Hosts: Raffi & Sreeja 📢 Coming in Episode 3: "Build AI Agents WITH AI" — we'll show how to generate the entire agent codebase using AI instead of writing it manually. Subscribe so you don't miss it!