The written version
Our heritage. In every language.
A sourced, interpretive translation of the entire Torah corpus, in each of the world’s major languages.
This is the same argument as the film above, with its sources, its figures and its limits attached, because an initiative about traceability should be traceable itself.
The five issues
5 issues
- 1 · Untranslated
- 2 · Mistranslated
- 3 · Not relatable
- 4 · Inconsistent
- 5 · No baseline
Many Torah texts remain completely untranslated.
Translate the entire corpus, with new translations where none exist.
Some texts are mistranslated.
Correct them, and name the source for every significant choice.
Some are translated correctly, but are not relatable to the reader. This is true of any reader, not only the young.
Make the meaning clear. Weave in the missing context, with every explanatory addition clearly marked.
Even when correct and relatable, translations lack consistency throughout the entire Torah corpus.
Carry reviewed decisions forward, while preserving differences in context.
And there is no authoritative benchmark translation that reflects the traditional Jewish reading of any Torah text.
Establish one: a baseline of accurate portrayal of Torah literature.
The vision
4 pillars
A sourced, interpretive translation of the entire Torah corpus, in each of the world’s major languages.
The aim is a baseline of accurate portrayal of Torah literature.
Sourced means every interpretive choice names the commentator or passage it follows. Interpretive means the meaning is made clear for today’s reader, with every explanatory addition marked as an addition. The example of Shir HaShirim 1:2 below shows both.
Until now, the number of accurately translated Torah books was limited. Soon, we will be able to open the whole library.
- For every seekerJewish or not makes no difference. People should be able to reach this knowledge, and whether they agree is up to them.
- Easy to findSearch engines and AI should lead every seeker to what traditional Judaism actually says. This is known as SEO and GEO, and it is a critical element of the initiative.
- Sourced and reviewedAI prepares the first draft; qualified scholars and editors decide. Every interpretive choice is visible and attributed.
- OpenThe aim is to digitize and open-source those translations, so that anyone seeking Jewish texts has access to authoritative translations rooted in traceable sources.
The TaryagAI Translation Initiative is grounded in traditional Judaic sources, with heavy emphasis on philology, intertextuality and historical context.
What is already happening
7 points
AI is already learning how to read our texts.
Today’s AI systems learn from content online, in digital libraries and databases: Jewish and non-Jewish, traditional and critical, careful and careless.Much of that material has never been checked.
Incorrect translations already exist online and in digital libraries. An AI system learns from all of it alike.Sounding right is not being right.
An AI system that has absorbed every translation it can find tends to average them. It cannot say which rendering follows Rashi and which follows a modern critical reading.Search for a Tanach translation online and the first result is often not one that reflects our tradition.
Reviewed Jewish sources need to be easier to find, and fixing that is part of what the TaryagAI Translation Initiative sets out to do.The Bible, including our Tanach, is already being translated at scale, and mostly not by the Jewish community.
People of other faiths intend to use AI to translate the Bible into every language by 2033: our precious Tanach, translated and interpreted through another faith’s lens.The Jewish world has already shown the method works.
In February 2026 Sefaria released the first comprehensive English translation of the Kli Yakar, drafted with AI and reviewed word by word by a human editor. But single volumes, each with its own conventions, do not add up to a corpus.Time is of the essence.
If an authoritative, tradition-aligned body of translation is not built now, other people’s translations will fill that space by default.
What is ours
This is about ownership: our heritage, our culture, our history, our legacy as a nation, and Hashem’s Torah.
It is our obligation and our duty to preserve the honor and truth of Hashem’s Torah. The aim is not to proselytize, but to take ownership of what is ours.
How the work is done
3 steps and an example
- Keep the source.The passage, the commentator, the interpretation, attached to the text and not to a memory of it.
- Record the decision.The wording, the reasoning, the review, and who made it.
- Build on it.Reuse the precedent where it applies. Record the contextual exception where it does not. Carry both to the next text.
AI prepares the first draft; qualified scholars and editors decide. Scholars review the wording and its basis; readers help test whether the meaning comes through.
יִשָּׁקֵנִי מִנְּשִׁיקוֹת פִּיהוּ
Close translation: “Let him kiss me with the kisses of his mouth.” Read alone, it sounds like a love poem.
Interpretive: [The Congregation of Israel, in exile:] “Let Him kiss me with the kisses of His mouth [let Him speak with us face to face once more, as at Sinai, out of His great love].”
The additions follow Rashi, Targum Shir HaShirim and Shir HaShirim Rabbah
Rashi
זֶה הַשִּׁיר אוֹמֶרֶת בְּפִיהָ בְּגָלוּתָהּ וּבְאַלְמְנוּתָהּ… אַךְ אֲנִי מִתְאַוָּה וְשׁוֹקֶקֶת לִהְיוֹתוֹ נוֹהֵג עִמִּי כַמִּנְהָג הָרִאשׁוֹן… כְּחָתָן אֶל כַּלָּה, פֶּה אֶל פֶּה
The Congregation of Israel speaks in exile, longing for the closeness of the first days, “as a groom to a bride, mouth to mouth.”
Targum Shir HaShirim
אַפִּין בְּאַפִּין כִּגְבַר דְּנַשֵּׁיק לְחַבְרֵיהּ מִן סְגִיאוּת חִבְּתָא
God spoke with Israel face to face, like one who kisses another out of great love.
Shir HaShirim Rabbah
רַבִּי יוֹחָנָן אָמַר: בְּסִינַי נֶאֶמְרָה
Rabbi Yochanan says the verse was said at Sinai.
The renderings here are a prepared demonstration and have not had independent scholarly or linguistic review.
The bracketed words are not in the pasuk. They are an interpretation, drawn from a named source, marked as an addition so a reader can see exactly where the tradition is speaking and exactly whose reading it is.
A call to join forces
No single institution can translate a whole library. Together, we can: one shared library, one heritage, one Torah, built together as one people.
- Yeshivot and seminaries
- Universities
- Publishers
- Libraries and archives
- Translators and editors
- Scholars and educators
Bring the texts. Bring the scholarship. Different strengths, a common purpose.
Let’s shape the standards and begin the work together.
Noam Harris · noam.p.harris@gmail.com · taryagai.com
This page makes no funding request. It is an invitation to join forces; the initiative’s funding plan is a separate document and is not changed by this presentation.
Who built it
Noam Harris is a professional translator and editor of classical Jewish texts, working in Hebrew, Aramaic and Arabic. His editorial and translation work includes the English Steinsaltz Tanya, and work with Koren Publishers and Sichos in English.
The publisher’s record
The publisher’s own sample lists him under both Translators and Editors: Tanya volume 1, page 5. No publisher endorsement of this initiative is implied.
What this page claims, and what it does not
9 limits
In short: TaryagAI is alpha-stage. The figures and assessments on this page are the initiative’s own and should be verified before formal use. The translated example is a prepared demonstration. The baseline is an aim, and no institution or publisher has joined yet.
Read all nine limits
- TaryagAI is alpha-stage. The public team-workflow demonstration uses a fictional team and prepared drafts; it calls no AI translation system. Fresh generation and editing are demonstrated separately in the local development application. Hosted feature parity and the effectiveness of the learning loop remain to be measured.
- The statement that AI systems learn from existing translations and commentary describes how today’s AI systems are generally trained on broad public text. It is not a claim about any particular system’s training data, and no claim is made that a particular future system will train on this initiative’s work.
- The Shir HaShirim 1:2 specimen is a prepared example chosen to show the method. It is not an approved edition and has had no independent scholarly review.
- The statements that many Torah texts remain untranslated, that some are mistranslated, that translations lack consistency across the corpus, and that no authoritative benchmark translation exists that reflects the traditional Jewish reading of any Torah text are the initiative’s own assessment. So are the statements that some translations have never been reviewed and that the first search result is often not one that reflects our tradition. All are stated without a prevalence statistic and are not the result of a published survey.
- The baseline is an aim. None of it has been published, and no claim is made that search engines or AI systems lead anyone to this initiative’s work today.
- The Why now figures — 2033 and the February 2026 Kli Yakar release — come from the initiative’s own sourced research notes and should be verified before use in a formal proposal.
- This page uses the narrower, checked wording: the Bible, including the Tanach, is being translated at scale — not the whole Jewish library.
- The call to join forces is an invitation. No institution or publisher has joined, committed or been approached through this page, and no partner institution or agreement has been established.
- No endorsement or partnership with Sefaria, Koren Publishers or any organization named in the initiative’s research is implied or claimed.
Sequence version 26, November 18, 2026. Version 25 with “No baseline” answered by “A baseline.” on the ledger slide, like the four issues before it. Version 25, November 18, 2026, was version 24 with the interpretive additions set upright rather than in italics, and the three sources arriving one at a time. Version 24, November 16, 2026, was version 23 with Rashi, the Targum and Shir HaShirim Rabbah each in its own block in the written example. Version 23, November 16, 2026, was version 22 with the written page in a bright style, and its sections and points opening and closing.
Earlier changes
Sequence version 22, November 16, 2026. Version 21 with the Hebrew set in Noto Sans Hebrew, for comparison with Heebo. Version 21, November 16, 2026, was version 20 with the Hebrew set in Heebo. Version 20, November 16, 2026, was version 19 with a new library line (“Until now, the number of accurately translated Torah books was limited. Soon, we will be able to open the whole library.”), Hebrew set in Frank Ruhl Libre, and the three sources quoted in their own Hebrew and Aramaic. Version 19, November 16, 2026, was version 18 with the sourced, interpretive line set apart on the opening screen and enlarged on the first slide, and a new slide before “Ask AI” on how AI learned from the translations and interpretations already online. Version 18, November 16, 2026, was version 17 without the four search screenshots (their lines now share one slide), and with a fuller sourced, interpretive example: Shir HaShirim 1:2 read with Rashi, the Targum and Shir HaShirim Rabbah, in place of Devarim 6:5. Version 17, November 16, 2026, was version 16 with four changes: one piece of music only, the film cut to about a minute and a half, a line on what translation opens for the next generation, and the sourced, interpretive translation stated first and repeated. Built from version 16, November 11, 2026: Version 15’s four changes, carried into the source: the fifth issue reworded, the heritage lines arriving phrase by phrase, a new hand-over between the two pieces of music (a short break, then the second piece begins with the vision), and four search screenshots on two slides, which hold a little longer. The full prose and the step-by-step example of one pasuk remain in version 10.TaryagAI home
Content drawn from TaryagAI Vision and Funding, the Personal Way In record, and the Edition 04 feedback audit, November 2026.