In my stage 3 task at HNG, I built Code Mentor, an AI-powered programming tutor that helps developers learn any technology through personalized guidance, project ideas, and step-by-step roadmaps — all powered by Mastra and integrated with Telex.im for real-time chat.

I wanted to create an agent that acts like a senior engineer who can: Analyze what you want to learn Generate a personalized learning roadmap Suggest projects based on your current level Encourage best practices and give feedback

Think of it as your AI coding buddy — available 24/7 to help you become a better developer.

Tech Stack: Mastra → For building and deploying the AI agent Gemini (Google) → As the underlying LLM Zod → For schema validation Telex.im → For real-time messaging and user interaction TypeScript / Node.js → Core language for logic

The Code Mentor agent was built using Mastra’s agent system. It has one main purpose, to understand what the user wants to learn and provide a personalized plan.

Telex.im Integration Once the agent was working locally, I deployed it to Mastra cloud and connected it to Telex.im. This allowed users to chat with Code Mentor directly inside the Telex interface.

Here’s how the connection works:

Telex.im sends a POST request to my Mastra Cloud endpoint whenever a user sends a message. Code Mentor processes the input and returns structured, human-like responses. Telex displays the response instantly. That simple connection made the agent feel alive, like chatting with a real developer mentor.

One major challenge I had was prompt refinement...I had to adjust the agent’s instructions multiple times to get consistent, “mentor-like” answers.

Building with Mastra made it incredibly easy to structure and deploy a powerful AI agent. If you’re working with AI or automation, this framework is worth checking out.