Classical Jewish texts on web and mobile, in Hebrew and English, for 100k+ readers. Lead engineer across the portal, the iOS and Android apps, the backend, and a GPT-4 translation pipeline for the French editions.
The Steinsaltz Center has been publishing Rabbi Adin Even-Israel Steinsaltz's translations and commentary out of Jerusalem since 1975, under the banner “Let my people know”. The platform is how that library reaches people now: a web portal and the Daily Study app on iOS and Android, carrying the canon with its commentary attached.
The portal. The tzedaka box is a first-class part of the product, not a donate link in the footer.
Hebrew and English in one layout, with notes and images anchored to the line they explain.
Daily Study: Daf Yomi and Daily Chumash as running cycles rather than a library to search.
What it carries
The canon, with the commentary attached. Talmud, Humash with Rashi, and Nakh in both Hebrew and English; Mishnah, Tanya and Rambam in Hebrew. Each text carries the Steinsaltz commentary, notes and images rather than sitting alone.
Daily Study rather than a library to navigate. Daf Yomi, Humash and Rambam arrive as the day's portion, which is how the cycles are actually kept.
Free core, premium library. Auto-renewing subscriptions open the full library and media, purchased and managed in-app on both stores.
A digital tzedaka box. Readers keep a pushka, set a goal, give in their local currency, and dedicate their learning in honour or memory of someone.
What I built
Lead engineer across the whole surface: the Next.js portal, one Flutter codebase for iOS and Android, and a FastAPI backend on GCP with Firebase.
Bilingual text rendering. Hebrew and English sit together, so right-to-left and left-to-right run in one layout with commentary, notes and images anchored to the lines they belong to. This is the part of the product that cannot be approximated.
Search built twice, once per language. English runs on semantic search: passages are embedded with an OpenAI embedding model and matched by meaning, so a question finds the sugya that answers it rather than the page that repeats its words. Hebrew runs on Algolia, because the language defeats naive embedding: unvocalised text, prefixes fused onto words and a root-based morphology reward a lexical index tuned for it over a vector one.
A GPT-4 translation pipeline for the French editions, reaching 99% accuracy so that editors proofread rather than translate from scratch.
Subscriptions and giving, spanning store billing on both platforms, local currency donations, sponsorship dedications and account-level management.
