---
title: "Slides: Recsys Keynote: Improving Recommendation Systems & Search in the Age of LLMs - Eugene Yan, Amazon"
category: "slides"
video_id: "2vlCqD6igVA"
sourceLabels: ["Public YouTube video frames", "Public YouTube metadata"]
---

# Slides: Recsys Keynote: Improving Recommendation Systems & Search in the Age of LLMs - Eugene Yan, Amazon

## Source Video
[Recsys Keynote: Improving Recommendation Systems & Search in the Age of LLMs - Eugene Yan, Amazon](https://www.youtube.com/watch?v=2vlCqD6igVA)

## 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/2vlCqD6igVA/slide-001.jpg]]

OCR text:

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

![[assets/slides/2vlCqD6igVA/slide-002.jpg]]

OCR text:

> Improving RecSys & Search
> with LLM techniques
> Al Engineer World's Fair 2025, RecSys Track
> with Latte & Mochi Sy
> | a Microsoft ary?

![[assets/slides/2vlCqD6igVA/slide-003.jpg]]

OCR text:

> Enriching exploratory search queries @ Spotify
> Extract from catalog titles,
> playlist names, podcasts
> Mine from search logs
> Use user's recent
> searches & saved items
> Apply metadata expansion
> rules (<artist> + "cover")
> Apply LLMs to generate
> natural language queries
> like Doc2Query, InPars
> Get regular,
> existing results
> Query rank optimized for
> stream, add-to-playlist, etc

![[assets/slides/2vlCqD6igVA/slide-004.jpg]]

OCR text:

> TL;DR:We buit a transformer-based
> works.
> Foryears,Stripe has been usngmachineleamingmodels trained on
> AIE
> dscretefeatures(BIN，p,paymentmetod,et.）to mprove our
> products forusers.And these feature-by-feature effortshave worked
> wll+15%coveron-30%fraud
> But these models have Imitations.Wehave to select and therefore
> Aside:Worked for
> constrain thefeatures considered bythemode.Andeachmodel
> equrestask-specifictrainingforauhorization,forfraud,fordsputes,
> Given theleamingpower ofgenerallzed transfomerarchitectures,we
> andsoon.
> paymentsandfraud
> obviousthati would-payments islikelanguagein some ways
> wonderedwhetheranLLM-stye approachcould workhere.t wasnt
> (structuralpattemssmlarto sytaxandsemantics,temporaly
> atStripetoo!
> knsywnainglent
> sequential)and extremely unlike language inothers (fewer distinct
> grammaticalrules).
> Sowebuitapaymentsfoundatonmodelsef-supevisednewok
> thatleamsdense,enerl-puposevctorsforeveytransactnmuch
> transactionsitdistillseachchargeskynasintoaingeeatl
> mbedsin.
> aws

## 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.