---
title: "Slides: Teaching Gemini to Speak YouTube: Adapting LLMs for Video Recommendations to 2B+DAU - Devansh Tandon"
category: "slides"
video_id: "LxQsQ3vZDqo"
sourceLabels: ["Public YouTube video frames", "Public YouTube metadata"]
---

# Slides: Teaching Gemini to Speak YouTube: Adapting LLMs for Video Recommendations to 2B+DAU - Devansh Tandon

## Source Video
[Teaching Gemini to Speak YouTube: Adapting LLMs for Video Recommendations to 2B+DAU - Devansh Tandon](https://www.youtube.com/watch?v=LxQsQ3vZDqo)

## Relationship To World's Fair 2026
These slides are extracted from a public AI Engineer YouTube video connected to World's Fair 2026. Speaker-matched clips are supporting context unless later confirmed as exact session recordings; official livestream recordings are day-level/event-level source material.

## Related Scheduled Sessions
- No individual scheduled session mapping has been assigned yet; treat this as an event livestream deck.

## Extracted Slides
![[assets/slides/LxQsQ3vZDqo/slide-001.jpg]]

OCR text:

> INNOVATIONPARTNER
> aws
> PLATINUMSPONSORS
> Graphite
> WWindsurf
> MongoDB
> daily
> augment code
> Workos

![[assets/slides/LxQsQ3vZDqo/slide-002.jpg]]

OCR text:

> AI Engineer
> World's Fair

![[assets/slides/LxQsQ3vZDqo/slide-003.jpg]]

OCR text:

> Recommendations drive majority of YouTube watchtime
> . o :
> ou a CHESSTIPS: &
> extaworo ‘ \ IFUL .
> Ga. ar Ce
> | A rr Others
> stele)
> Ts
> = a A
> - *
> fo —_
> Home WatchNext Shorts Search
> a a Microsoft ary?

![[assets/slides/LxQsQ3vZDqo/slide-004.jpg]]

OCR text:

> Personalized Recommendation Problem
> =: =
> fa peers :
> f ( 7a _—) ) ~ FF
> Ad ’ [ Pm
> a
> === :
> User Context Recs
> a aws
> a)

![[assets/slides/LxQsQ3vZDqo/slide-005.jpg]]

OCR text:

> LRM: Adapting Gemini for Recommendation Tasks
> . Home
> > mae, » WatchNext
> Gemini | AX a “silos
> a ’ > — Search
> + -
> r r 24 - Music
> -m ee Ak
> soe ES a
> vid
> 3 Voulube LRM . Ranking » Experiments
> Base Gemini checkpoint, Aligned to Served across
> adapted for YouTube tasks surfaces
> tions
> a
> me V1 cai)
> a a Microsoft =a)

![[assets/slides/LxQsQ3vZDqo/slide-006.jpg]]

OCR text:

> . . °
> LRM finds unique, personalized recs for our hardest recs problems
> User demo ete = aa ay
> peers _
> 24 yt old. Female . - % a
> US Android cay oO avon -
> Smee s 7 Te
> Ls oe PPro me ed _ rae ~ c
> Watch history “nee eee , ser (re aa =
> a ts a) DB cota arn
> 3 a bo rd ba Cherrgeera
> 2 es
> @5-5=° @ LT Ont WatchNext: men’s track races. no women
> LRM prompt . .
> F Lc
> : US 24 years femele a , . 7 x bad
> : - | ; a ba S a
> : : 1 : — a
> -. WHAT A COMEBACK’ { Men's 400m | i JN ve ca ad
> BParis2024 highlights 4 we... ep ae isu 5 a Ss se noi 5
> wpe pat @ yoy Mick wegen Leone wt e US cram Go warmreny ecrnatie
> Match feet res * UNCATCHABLE © mortd record 400m 7 40800 -viay fe Boma Orpigpe meat
>  \ Paylor Swift!” boot Partin Pare Copenoe foe cron
> 3 Lo Boe LRM: finds related women's races
> Qo
> | a Microsoft §=ooou®

## Slide-Derived Subjects To Review
Subject extraction uses video title, related session titles/descriptions, transcript context, and OCR text when available. OCR is best-effort and should be reviewed against the embedded slide images.