Sh’elah — Devlog 4

Over the past week or so, I had…

• Fixed an issue with RLS in Supabase so that one user doesn’t see another user’s personal info, as well as security issues with Vercel and Clerk (Which was a pain to fix :<)
• Found out my login system and database weren’t linked to each other correctly, which was breaking authenticated reads/writes, and fixed it
• Added an ai usage tracker and fixed old usage records so the spending log is correct, and so I don’t go broke when other people use my site’s ai
• My webhook (discord error bot) was malfunctioning (and still kinda is in clerk :<) and no longer is! YAY MORE ERRORS FOR ME! (at least I’ll know why they happen :|)
• Fixed a bug where the site was incorrectly caching files in browsers, making things slower and more expensive (yay less money to spend!!!!!)
• Ran accessibility tests and fixed any issues they found, and made my site more ready for releasing v1.00!
• Removed old files
• Fixed some legal documentation issues (like my Privacy Policy and ToS) so I don’t get sued!!!

Let me know if there is anything y’all wanna see on my site in the future
p.s. I’m planning a huge commit soon announcing version 1.00 of my website so if you haven’t seen my first devlog announcing that, go check it out.

Sh’elah — Devlog 3

Over the past week or so, I had…

• Fixed a bug where a malformed cache config could crash the whole app on startup.

• Split a file that handled AI question-answering into smaller, cleaner, more compact files, and added tests to make sure that both versions (one for gemini and one for claude) always answered the same way with the same guardrails.

• Sped up busy/resource heavy pages (library, prayers, texts, daily study, and the calendar lookup) by adding a shared caching layer, so they don’t request the database on every single request anymore.

• Added rate limiting across the whole site instead of having it scattered in different places.

• Cleaned up a batch of files that were referenced in the code which was uncommitted and caused boot errors.

• Simplified a few functions that had been made too complex to safely maintain.

• Sped up cold starts by delaying AI library imports until they’re actually needed.

Let me know if there is anything y’all wanna see on my site in the future
p.s. I’m planning a huge commit soon announcing version 1.00 of my website so if you haven’t seen my first devlog announcing that, go check it out.

Sh’elah — Devlog 2

Over the past 9 hours or so, I had…

• Removed a leftover routing rule in Vercel that was returning a 404 error in my production site.

• Added a server-side daily budget per user, so if someone goes past it and heavily uses the AI past what’s normal, they get rate limited and can’t spam multiple requests at once. Also fixed an issue where AI usage was being reported as $0 when it wasn’t.

• Updated my site’s README.md on GitHub, fixed a license mismatch, and added a CONTRIBUTING.md and a code of conduct.

• Old logs and history get deleted on a schedule now, so trash doesn’t build up over time.

• Built a dark-mode accessibility checker and fixed contrast issues. Also integrated circuit breakers so when an external resource fails the site doesn’t crash.

• Added About/Help/glossary pages, a sitemap, a feedback widget, and analytics that you can opt into.

• Added CI checks and Vercel Speed Insights to track loading times.

• Swapped the GPS permission for zmanim (prayer times) based on location to cookie/IP-based location instead.

Let me know if there is anything y’all wanna see on my site in the future

p.s. I’m planning a huge commit soon announcing version 1.00 of my website so if you haven’t seen my first devlog announcing that, go check it out.

Sh’elah — First Devlog:

I’ve been working on Sh’elah for a while now and finally uploaded it here. It’s a Jewish learning app that combines a searchable Torah library (similar to Sefaria) with an AI tool designed to actually answer your halacha questions. Instead of just throwing related sources at you that don’t end up answering the question, it walks you through the actual answer, explains the reasoning, cites the sources, and tailors everything based on your community’s specific customs.

What it does today

• Answers questions across any area of Jewish law using primary sources rather than generic guesses.
• Allows you to search the library with customs broken out by community (Ashkenazi, Sefardi, Teimani, and others).
• Calculates live zmanim (prayer times) and a Hebrew calendar based on your exact location.
• Ends every answer with a clear reminder: this is for learning purposes, not an official ruling, and real questions should always go to your rabbi.

It works right now, but it’s still in the early stages, so I want to be upfront about what still needs work before calling it a complete release.

Making the AI reach for things, not just recite them

• Right now, the AI only knows what gets put into the prompt ahead of time. I am working on moving it toward real tool use so it can pull zmanim, calendar dates, commentaries, and community customs dynamically as needed.
• Judaic texts and laws will always remain the main priority, with web searching kept strictly as a last resort.

Getting the safety layer right (in progress)

• Halacha touches sensitive material, and I want to make sure a younger user doesn’t receive graphic content when the real answer should simply be “ask your rabbi for details.”
• Any queries related to medical issues, mental health, or abuse will automatically route to real professional resources rather than generating an AI ruling.

Behind-the-scenes technical updates

• Breaking down a backend file that got way too large so the code is easier to maintain and debug.
• Securing the caching layer so it handles concurrent load safely.
• Adding circuit breakers to prevent a third-party outage from crashing the whole platform.

What needs to happen before full launch

• Verifying every feature works properly while keeping the user experience smooth and simple.
• Completing a thorough security review (already finished).
• Setting up error tracking and rate limiting to protect the site from abuse.
• Performing a full accessibility audit.

Building this isn’t going to be easy, and the order of these steps will probably shift around. However, this is the general direction of the project, and I’ll share more updates as new features roll out.