Tarek Gara.

· talks

Talks.

Topics I'm available to speak on, with rough abstracts. None of these are off-the-shelf — shaped to fit the audience and venue.

Keynotes, workshops, podcast appearances, panel moderation. Email hello@tarekgara.com to book; bio and headshot on request.

Topics

four talks, drawn from the practice
  1. Where alignment work breaks down in Arabic

    For AI safety teams · multilingual NLP labs · enterprise GenAI buyers

    What multilingual evaluation surfaces when you stop treating English as the baseline. Idiom and register that slip past safety filters, test sets that weren't drawn for non-Latin scripts, the operational shape of language-specific QC. Drawn from leading multilingual AI quality at Innodata across Arabic, Hebrew, and English; concrete examples, not abstractions.

    Format 30–45 min talk · or 90 min workshop with a hands-on eval segment

  2. Red-teaming in non-English: a beginner's notebook

    For AI safety conferences · policy circles · researchers entering the field

    I'm learning red-teaming in Arabic in public, via Thagrah. Honest survey of what transfers from English literature, what doesn't, what tooling actually exists, what the obvious gaps are. Less a master class than a working document — useful precisely because the path is still being drawn.

    Format 25–35 min talk · or a fireside conversation

  3. Shipping small focused tools fast

    For founders · indie builders · product teams · engineering meetups

    What I've learned shipping Folio, Tayyar, and Thagrah on tight clocks. The scoping discipline, the cuts that hurt, the difference between iterating in public and dithering in public, what shipped and what didn't.

    Format 30–45 min talk · or 2-hour build-along workshop

  4. The political life of digital infrastructure

    For AI governance forums · cyber-policy events · academic seminars

    What the MA in Cyber Politics & Government at TAU looks at, and why a cyber-politics frame matters for AI governance specifically — including the gap between how alignment teams think about safety and how policy makers think about risk.

    Format 35–50 min talk · or seminar

How I work

a few things worth knowing
  • Audience-shaped. Every talk gets rewritten for the room. The abstracts above are the bones; the muscle is built for the venue.
  • Honest about the edges. I'll tell the audience where I know enough to teach and where I'm still learning. The latter is sometimes the more useful part.
  • Multilingual when it helps. Comfortable in English, Arabic, and Hebrew. Happy to do mixed-language Q&A.
  • Quiet on stage. I'm a writer first, which means measured pacing and a heavy reliance on the slides carrying their weight. No theatrics.

Bio

paste-ready, two lengths

Take whichever fits. Headshot at /me.jpg (square crop). Edit freely; just keep the facts straight.

Short

≈ 50 words

Tarek Gara leads multilingual AI quality at Innodata, across Arabic, Hebrew, and English. Founder of Folio, Itaha, and Mawrid Research. Author of four books, including Mortal Past and Nonperson. Finishing an MA in Cyber Politics & Government at Tel Aviv University. From and in Jatt.

Longer

≈ 150 words

Tarek Gara leads multilingual AI quality at Innodata, where he audits and improves model outputs across Arabic, Hebrew, and English. His current focus is the gap in non-English alignment: prompts that English models handle cleanly but slip past guardrails in Arabic. He is the founder of Folio (an integrity-first research workspace for academics), Itaha (a non-profit advancing digital accessibility in the Arab community in Israel), and Mawrid Research (an editorial studio for cross-language academic work). His tools in development include Tayyar, a political-compass map of MENA actors, and Thagrah, an open evaluation harness for adversarial prompts in Arabic. He is the author of four books: Mortal Past (2020), The Flag Bearers (2021), Crescent Moons (2023), and Nonperson (2024). He is finishing an MA in Cyber Politics & Government at Tel Aviv University. From and in Jatt.