The Megillah App Sprint, Purim 2026
How shluchim in the AI for Shlichus group shaped a synchronized Megillah app in about ten days before Purim 2026, and the lessons from that sprint and from an Esther reader that broke on Purim day.
On this page
This page is history: what happened in the ten days before Purim 2026, when one shliach posted a Megillah app and the group turned it into a shared project. The app as it is today is on megillah.app.
Timeline
Dates are from the chat. Purim 5786 began on the evening of March 2, 2026.
| Date | What happened |
|---|---|
| Feb 20 | A shliach posts megillah.app, built with Claude. The first default reading time is 35 minutes; members call it long. |
| Feb 21 | The code goes public on GitHub. Translations in six languages, a sound pause button, a shake-to-gragger setting and a link to Chabad.org are added. |
| Feb 22 | Before dawn, a member posts "Shluchim Vibe Coding Sprint: Purim is in 10 days" with an idea list: translations, voice recognition, a kids version, QR sessions, sound effects, word highlighting, synchronized shul reading. The live mode ships the same day, and an updates page goes up. |
| Feb 22 | The group admin opens a separate AIFS Builders group for vibe coding and agent building, after a poll, and asks that Megillah app talk move there. |
| Feb 22 to 23 | Members object to the 1917 JPS English from Sefaria as archaic. Shluchim testing in their shuls report lag between phones. |
| Feb 25 | Broadcaster sign-in is clarified: open /live and enter the session ID and password. |
| Feb 25 to 26 | A large update: verse-by-verse view by default, centered auto-scroll, smaller default font, transliteration and translation of the brachos, translation settings in the top bar, broadcaster-set defaults, an "unfollow" button. The English moves to Metzudah, then to the Kehot translation once Kehot gives the go-ahead. |
| Mar 2 | Illustrations from TorahApp.org are added. COLlive publishes "700 Shluchim Built a Megillah App in 10 Days." |
| Mar 2 to 3 | Purim. Chabad Houses use it in shul. |
| Mar 3 to 5 | Debrief and feedback; a parallel Esther reader shares a frank post-mortem (below). |
Feature requests, granted and turned down
| Request | Outcome |
|---|---|
| Several languages | Shipped (six at first, ten by the end of the season) |
| Transliteration | Shipped for the verses read aloud and the sons of Haman |
| Pause, font and size controls | Shipped |
| Haman sound effects and shake-to-gragger | Shipped (today the site uses tap-Haman) |
| Live broadcast to every phone in shul | Shipped |
| Images | Shipped as an optional layer |
| Auto-scroll by voice recognition | Turned down: "the tech is not ripe enough," too much delay |
| Turning on Do Not Disturb automatically | Not possible from a website; would need a native app |
| Settings that match between broadcaster and followers | Promised as a fix during the sprint; broadcaster defaults now carry over to new joiners |
The other Megillah reader: a post-mortem
In the same weeks a member launched an image-rich Esther reader at learningtanach.org. It worked on Purim night. On Purim day, after a round of fixes, it failed. The builder posted an open post-mortem on March 4 and 5, 2026. In his words, he "kept adding pieces using AI tools," and each addition made the code more fragile until there were "six different timing systems contradicting each other." He offered refunds to people who had paid for premium features and rebuilt the reader from scratch. The site still drew about 7,000 visitors in two days, and the images alone took more than 100 hours.
Lessons the group drew
These come from the post-mortem and the debrief that followed.
- Freeze what works before the big day. If people already like version 1, do not ship a stack of fixes the night before. The learningtanach.org reader broke after fixes, not before them.
- Keep versions you can roll back to. A member's advice after Purim: number your releases (1.0, 1.1) so a bad update can be undone in minutes.
- Each AI edit only sees the code in front of it. Many small AI patches can each look right and still leave the whole app inconsistent. When it starts to wobble, ask the AI to review the whole design, or start clean.
- AI does not replace the full software process. Testing on real phones in a real shul, a plan, and someone who reads the code still matter.
- Ship where it counts, then stop. The Megillah app worked because it was short, one-time-use and needed no accounts. In October 2026 a builder pointed out that long-running apps that shluchim use all year are much harder; the Megillah sprint is not proof that a group can maintain a full Chabad House system the same way.
- Plan for one maintainer. Nearly all merging went through one person. Agree in advance who can take over if that person is busy on Purim.
Why it is remembered
The sprint showed a few hundred shluchim, most of them not programmers, that a free tool for their shul could go from idea to Purim in ten days when it stays inside the vibe-coding rules: no logins, no payments, no database of people. One participant reported the reaction after Purim: "People were telling me Chabad are geniuses and leaders. People loved it."
It also set the pattern the builders group still uses: post a working version, take requests from the chat, and ship changes within hours.