When we started running our courses, we shared a Google Doc with participants containing links to required readings. Facilitators would use a separate Google Doc with discussion prompts to engage their cohorts of four to six participants.

Once we saw a real appetite for our courses, we started hosting the curriculum on a static [Squarespace website](https://web.archive.org/web/20220908073448/https://www.agisafetyfundamentals.com/ai-alignment-curriculum), which looked more professional and made the course more discoverable.

<Embed url="https://bluedot.org/u/2024/10/Untitled-design.png" caption="Our old website, still viewable on the Wayback Machine!" />
_Our old website, still viewable on the Wayback Machine!_

However, if we wanted to support ten times more people in pursuing a high-impact career, we'd need to overcome three main issues with our setup:

1. **We had no idea who was doing the readings** or if participants found them useful. We relied on direct feedback from facilitators and participants.
2. **We spent hours each day sorting participant requests,** particularly when they asked to switch sessions or needed help finding meeting links.
3. **We had to manually update the website** with curriculum changes as we solicited feedback in a Google Doc.

How could we collect data on where to improve the course and free up team time to do so?

## Standing on the shoulders of giants

Our problem wasn't unique. Hundreds of thousands of people design and deliver educational content. For example:

- Some universities use [Moodle](https://moodle.org/) for digital learning.
- Minerva University built [Forum™](https://www.minervaproject.com/platform), a hybrid platform for curriculum design and learning engagement.
- Independent educators host massive open online courses (MOOCs) on platforms like [Coursera](https://www.coursera.org/) or [Udemy](https://www.udemy.com/).

Surely, we could find an off-the-shelf platform that works for us… right?

## Our learning platform requirements

From our experience running the course, we identified several key features for a learning platform:

- Displaying the content (e.g. session overview, required and optional readings, resource guidance)
- Engagement tracking (e.g. reading completions, time spent preparing)
- Mechanisms for soliciting participant feedback
- Integration with Airtable for real-time curriculum updates
- Personalised session information for participants, including the meeting link
- Access for facilitators to view their participants' responses to polls and open questions

## Checking out what's on-the-shelf

We explored various learning platform options, including [Brightspace](https://try.d2l.com/en-eu/brightspace/he/), [360 Learning](https://360learning.com/), and [Coursemap.ai](http://Coursemap.ai). We even tried to ask Minerva if we could use Forum, but no dice!

The most promising option was [Thinkific](https://www.thinkific.com/features/create-digital-products/online-course-platform/).

> Thinkific is a software platform that enables entrepreneurs to **create, market, sell, and deliver** their own online courses.

With its clean user interface and built-in activities, Thinkific seemed like the right fit. Here's a [demo](https://app.usebubbles.com/aWDQYZjMqRyFZMWbK6Ajk2) of the first week of the AI Alignment course built on Thinkific.

However, after building a quick MVP, we discovered several key limitations:

1. **Engagement data was limited and hard to access.** Instead of having a real-time view of engagement, we would have to manually download spreadsheets.
2. **Users had to go through content in order**, but some participants preferred jumping around.
3. **We couldn't group content by session**, requiring a new 'course' for each session.
4. **No integration with Airtable**, meaning we'd still have to make manual curriculum updates.

Using off-the-shelf platforms like Thinkific often means sacrificing flexibility for ease. These learning platforms were built for a specific customer and if you didn't fit the profile, you would have to adapt your course to the platform or come up with a complicated workaround.

It's like choosing between buying clothes off the rack or sewing your own. The more customisation or better fit you want, the less suitable off-the-shelf products become.

But as with clothes, there are places you can go to get a more custom-made option before you pull out the sewing machine.

## Custom no-code solutions

We needed a solution that offered flexibility without extensive coding. As a small team of four (now six!), we wanted to avoid using up our capacity on building software.

The middle ground was using a no-code tool like [Softr](https://www.softr.io/), [Glide,](https://www.glideapps.com/) or [Appy Pie](https://www.appypie.com/). We ultimately chose [Bubble](https://bubble.io/), a full-stack no-code editor.

Bubble enabled one full-time staff member without a software engineering background — me! — to build a custom learning platform.

In November 2022, about a month after joining the team, I started building what we called our Learning Management System (LMS). A few weeks later, we rolled out the MVP, and within two months, we fully migrated our curriculum from the website to the LMS.

Bubble enabled us to:

- Design, build, and test every aspect of participant experience on the platform
- Automatically display live updates to our curriculum from Airtable
- Show personalised information to each user
- Analyse participant engagement with course content (by syncing data back to Airtable)
- Allow participants to switch sessions with minimal intervention
- Allow facilitators to update session information with minimal intervention

## Looking back

If we hadn't decided to invest in building our own learning platform, which we now call the [Course Hub](https://course.aisafetyfundamentals.com/alignment), I don't think we would have gone with an off-the-shelf solution unless we found something that _really worked_. We wouldn't have tried to force the form of our courses against their function.

Instead, we might have attempted to build standalone tools to connect our website with Airtable, process cohort switching and track participant engagement. However, we would have been wary of overwhelming our participants with too many different software programs, given that they already had to juggle Zoom, Google Docs, and Slack.

Even if Minerva allowed us to use Forum, we would have quickly realised it wasn't the right platform for us because it didn't allow for the granularity of data collection we wanted or the speed of changes to constantly improve our participants' course experience.

There are times when it makes sense not to recreate the wheel. If there had been an off-the-shelf Course Hub, we wouldn't have built our own. But in the absence of a good solution, sometimes you've just got to do it yourself.


Why we spent 3 months building our own learning platform instead of using one off-the-shelf

Li-Lian Ang • October 2, 2024