Week in Review: 110 Commits, a Community Platform, and a Video Toolkit
Built a Skool-style community for Firstly Academy, started a video toolkit, added Amplitude analytics, and somehow hit 110 commits. This is the new normal apparently.
I looked at my commit history this week and genuinely laughed. 110 commits across two projects. That can’t be right, surely?
It is. Tbh it’s quite disorientating.
Here’s what actually happened.
Monday–Tuesday: Community Platform
My Firstly Academy students needed somewhere to practise English together outside of lessons. Forums feel dated. Discord is too chaotic for language learners. Skool has the right vibe but costs money and doesn’t integrate with anything.
So I built one.
Two days later: posts with infinite scroll, threaded comments, reactions (like, celebrate, helpful, funny), leaderboards, daily challenges, and member profiles with activity feeds.
The gamification ties into the main app’s XP system. Do exercises, earn XP. Post in the community, earn XP. It all feeds the same leaderboard. Students can see their streaks, their level, where they rank.
From database schema to production in two days. I’m still not entirely sure how that happened.
Wednesday–Thursday: Notifications and Polish
A community without notifications is just a graveyard you have to manually check. So I added:
- Notification bell with unread count
- @mentions that actually notify people
- Reaction batching (so you don’t get 15 notifications when people react to your post)
- Streak tracking with emoji indicators
- YouTube video embedding in posts
- Image uploads for teachers
The small things that make an MVP feel less like an MVP.
Amplitude for English-Exercises.org
Also added Amplitude analytics to english-exercises.org this week. Thanks to Molly Cantillon for the recommendation.
I wanted to understand where students actually drop off. Do they finish exercises? Do they give up halfway through? Which question types cause the most problems?
Now I can see:
- How far users get through each exercise set
- Where they abandon
- Which grammar points have the highest completion rates
- Time spent per question type
This feeds back into exercise design. If everyone’s dropping off at question 8, something’s wrong with question 8. If cloze exercises have 90% completion but transformation exercises have 40%, that tells me something about difficulty calibration.
Data-informed teaching materials. Novel concept.
Friday: New Project
Started something completely different: an ESL video toolkit.
I record video lessons. The editing workflow is painful. So I’m building tools to make it less painful.
The killer feature: WhisperX integration with word-level timestamps. Not “close enough” caption timing. Precise forced-alignment timing that matches exactly when I say each word.
Also added:
- Multi-brand support (different intro/outro styles for different platforms)
- Short-form exports for TikTok, Reels, Shorts
- Vertical video cropping that actually works
It’s scratching my own itch. If other ESL teachers find it useful, bonus.
Saturday: Marketing Agents
This one’s a bit meta. I built a team of AI agents to help with marketing.
- Growth marketing: SEO, ads, referrals
- Copywriting: landing pages, emails, UX writing
- Content marketing: blog, YouTube, social
- Coordinator: ties them together
They can invoke each other. Ask the coordinator for a campaign strategy and it’ll pull in the specialists as needed.
Is this over-engineered? Probably. Is it fun to talk to a virtual marketing team? Absolutely.
The Pace
110 commits in a week. Community platform from scratch. Notification system. Video toolkit. Analytics integration. Marketing agents.
This pace comes from three things:
- Knowing the domain deeply – I teach ESL, I know what students need
- Having more ideas than I know what to do with – the bottleneck is execution, not ideation
- Using Claude Code like my life depends on it – I direct, it executes
The AI doesn’t decide what to build. It doesn’t understand why a community platform matters for language learners or why caption timing needs to be word-level precise. That’s the human part.
But once I know what I want, shipping it is just… fast now.
It’s still weird. I keep expecting to hit a wall where this stops working. So far, no wall.
What’s Next
- More grammar modules for english-exercises.org (informed by the new analytics)
- Community platform refinement based on student feedback
- Video toolkit MVP that I can actually use for my next batch of lessons
Building in public continues.
Follow the build: X @charliecrowley
Related: