Augmenting Education with Machine Learning

This rough neural network proof of concept is for a letter formation gap analysis, learning, and screening framework. I dream that it could one day aid students and educators.

Reversing characters (e.g. bd, pq) is a very common struggle for children learning to write. While normal, it can also be indicative of an underlying visual processing or learning disability. This small demo attempts to perceive letter reversals and case. See below for further details.

Instructions: This demo expects you to write a letter (a-Z) in the box matching the letter goal. Press the ‘Clear’ button and try writing something. Use the ‘Change Letter’ button for a different letter goal. A few examples are provided in the drop down.

Letter Goal: e


Correctness Tolerance
Allow Reversals
Case Insensitive

Result Remarks
Loading ...

A Personal Perspective

During the pandemic when kids were sent home, I manually performed exactly what this neural network and framework is attempting to accomplish with my Son.

From that experience, an interesting scale and optimization challenge presented itself. An educator couldn’t possibly give individualized, highly focused, and synchronous attention to 30 students in a classroom simultaneously.

It struck me that was this was the perfect challenge for neural networks to solve. After conducting extremely thorough research, I discovered something highly logical—Vulcans were using machine learning to solve education scaling in the year 2286 :)

The Learners Perspective

It can be very frustrating to be right and wrong at the same time. My goal is to provide a practice and learning environment with a positive feedback loop—a correct but reversed character needn’t be wrong! Combined with immediate feedback and guidance, that is individually tailored and automated, learners are setup for success.

Educators and Parents

This framework is capable of measuring and quantifying behind the scenes, with an overarching goal of driving impact. The collected analytics would be useful on a variety of fronts:

  • Tailoring and optimizing learning experiences in the moment.
  • Measuring individual improvement and detecting regressions over time.
  • Age adjusted percentile charting for IEP placement and tracking.
  • Aiding the strategic allocation of additional resources and services for students.
  • Using automation to regularly/efficiently/scalably screen many students in parallel.
  • Most importantly—maximizing outcomes for students and helping our educators.

Learning Character Formation

One key character reversal mitigation strategy is learning the proper starting point and writing flow for each character. Typically character forming is taught in groups with other similar forming characters.

I’m working on a neural network variant capable of detecting and scoring these formations. Beyond an additional metric, the idea is to detect and create an automated in place learning and reinforcement opportunity—without requiring direct manual observation.

On Numbers

Though fully stubbed out, numbers are not currently supported. I’ve focused on letters as it is the larger technical challenge at hand. Further along numbers will be feathered in.