Mentra

Inspiration

A college degree and job security become more loosely correlated every year. One thing that still differentiates workers in all fields is hands on skills, and these can only be learned by doing. Many AI services have tried and failed to personalize learning, by either failing to innovate past chatbots, or simply lacking the long-term context to teach a long-term skill. Meet Mentra.

What it does

Mentra introduces a new approach to learning, where you are learn by doing, directly within your software environment; no context switching necessary. An agentic lesson planner curates lessons, with specificity down to each click. The user is then guided through the curriculum by a translucent text overlay placed on their screen—an agentic mentor challenging and guiding on them through hands-on experience.

How we built it

When discussing on the best way to build Mentra, phrases like AI agents, long context, and AI workflows were brought up then and again. The core problem we wanted to solve with Mentra was to reduce the context switching a user has to make when they have software in one screen, and a lesson in another. By solving this, we also open the door to providing instant feedback relevant to the context at that point in time. With all these goals and ideas in mind, we decided to build Mentra as an AI workflow, window-embedded lecture guide that uses AI agents with long memory to teach software to people more effectively.

Challenges we ran into

One big challenge that nearly killed our idea was configuring agentic flows for multi-step instructions including feedback to the user when they make a mistak, all while keeping AI context drift to zero. We solved this by segmenting Lectures into multiple Steps. The AI only needs two things in its co text at any point in time; the overarching goal of the lesson, and the goal of the specific step (could be a step to fix mistakes from a previous step). We saw massive improvements in AI coherence which is paramount to any educational endeavour.

Accomplishments that we're proud of

The small translucent box at the top right corner that shows instructions to the user is not only innovative, but incredibly helpful. We’re proud of solving window-embedded UIs for instructive feedback.

What we learned

Start out with the simplest way to solve a specific problem. Complexity should increase out of necessity, not because you want flashy features.

What's next for Mentra

We believe enterprise will benefit the most from Mentra. Imagine a world where every employee is a rockstar at each software they learn. Imagine the productivity boost!