How Omie Picks Today's Lesson for You
- The inputs
- The library side
- The selection logic, simplified
- What learning-style preference does
People often ask how Omie decides which lesson to deliver each day. It's a valid question, especially in an age where many learning platforms claim to harness the power of AI without providing clarity on how it works. At Omie, we prioritize transparency and simplicity in our recommendation process. Let's delve into the specifics of how we select your daily lesson, stripping away the jargon to reveal the straightforward mechanics behind the scenes.
The Inputs
When you first sign up for Omie, we gather six essential pieces of information about you:
- Your role: This helps us understand the context of your work.
- Two to three goals: You articulate these in your own words, giving us insight into what you want to achieve.
- Your seniority: Knowing your experience level allows us to tailor the complexity of the content.
- Learning-style preference: Although we don’t overly rely on this, it gives users a sense of personalization.
- Time zone: This ensures lessons arrive at a convenient time for you.
- Skills you've touched: If you've engaged with specific skills, that data informs our recommendations.
These inputs are intentionally minimal. Lengthy onboarding forms can deter users, and many additional inputs would yield only marginal benefits in predicting your learning needs. The real magic happens after you've been using Omie for a week, during which we start to build a more comprehensive picture of your preferences and behaviors.
The Library Side
On the other end of our system lies a robust library of lessons, each meticulously tagged and categorized. Every lesson is associated with:
- Vector embeddings: These capture the content's intricacies.
- Skill tags: Each lesson is labeled with relevant skills for easier filtering.
- Difficulty ratings: This ensures that the lessons match your current level of understanding.
- Estimated read times: This helps you manage your learning in bite-sized chunks.
- Prerequisites: These indicate what knowledge you should ideally possess before diving into a lesson.
- Follow-on lessons: This guides you toward related content, enhancing your learning journey.
While some of this metadata is curated by humans, a significant portion is generated by our AI, Claude. However, we don’t blindly trust automated tagging. We monitor the confidence scores of the tags, ensuring that only the most relevant ones are considered in our recommendations.
The Selection Logic, Simplified
Every morning at 6 a.m. in your local time zone, our selection process kicks off. Here’s how it works, broken down into digestible steps:
-
Goal Fit Filtering: We narrow down the library to a few hundred lessons that align with your stated goals.
-
Spaced Repetition: Recently viewed lessons are pushed down the list, while topics covering the same skill as a lesson you encountered two days ago are elevated. This method leverages the spacing effect to enhance retention.
-
Difficulty Curve Application: Early on, we provide foundational lessons. As you engage more, we introduce harder and more complex topics, ensuring you’re constantly challenged without being overwhelmed.
-
Dislike Filtering: If you've marked a lesson as “not for me,” it gets filtered out within a defined similarity radius for 30 days. We respect your preferences.
-
Fit Score Ranking: The remaining lessons are ranked based on a fit score that considers embedding similarity, your completion rates, and user ratings from similar profiles.
-
Sampling, Not Argmax: We avoid the pitfall of sending you only the top-ranked lesson. Instead, we sample from the top choices to prevent your learning experience from becoming stagnant.
-
A Touch of Taste: Finally, a small fraction of lessons is designated as editor picks. These may not be the most obvious choices but are deemed valuable for your role, adding an element of surprise to your learning experience.
What Learning-Style Preference Does
You might wonder about the role of learning-style preferences in this process. While we collect this information to make users feel heard, the reality is that it has minimal impact on the overall effectiveness of our recommendations. Research shows that learning styles have little effect on retention. Therefore, while we may slightly re-rank lessons based on whether you're more "story-led" or "framework-led," the difference is negligible.
What We Don’t Do
It's crucial to clarify what our system is not. We do not rely on a black-box deep learning model that makes opaque decisions about your lessons. Instead, our recommendation pipeline is built on clear rules, a fit score, and a sampling strategy. Each step is transparent, inspectable, and reproducible. If you ever question why you received a particular lesson, we can provide a straightforward answer.
A Practical Example
Imagine you are a mid-level project manager aiming to enhance your leadership skills. You've stated this goal during onboarding, and your profile indicates you've previously engaged with lessons on communication and team dynamics. After your first week of using Omie, we notice that you frequently open lessons related to leadership but skip those focused on technical skills.
Every morning, our system will filter out lessons that don’t align with your goals. It pushes down topics you've recently viewed while bumping up lessons that expand on the leadership skills you've interacted with before. Perhaps one morning, you receive a lesson titled, "Leading Through Change." This lesson not only fits your goal but also introduces a new concept you haven’t encountered recently, keeping your learning fresh and engaging.
Conclusion
In summary, Omie's lesson selection process is straightforward and rooted in real data rather than mystical algorithms. By combining your inputs, your engagement signals, and a rich library of well-tagged lessons, we deliver personalized learning experiences that evolve with you.
Curious to see what Omie picks for you tomorrow? Take the Omie Skill Assessment and discover your tailored learning path!