---
title: "Reconstructed Slides: GraphRAG: The Marriage of Knowledge Graphs and RAG: Emil Eifrem"
category: "slides"
video_id: "knDDGYHnnSI"
sourceLabels: ["Cropped public YouTube video frames", "Local OpenCV slide-region detection", "Local RapidOCR"]
---

# Reconstructed Slides: GraphRAG: The Marriage of Knowledge Graphs and RAG: Emil Eifrem

## Source Video
[GraphRAG: The Marriage of Knowledge Graphs and RAG: Emil Eifrem](https://www.youtube.com/watch?v=knDDGYHnnSI)

## Method
This deck is reconstructed from the existing video frame captures by detecting likely slide regions with OpenCV, cropping/upscaling those regions, deduplicating similar crops, and OCRing the cropped slide images locally. It is a cleaner companion to the full-stage frame deck.

## Reconstructed Slides
![[assets/reconstructed-slides/knDDGYHnnSI/slide-001.jpg]]

- Source frame: `slide-001.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `158.63`

OCR text:

> INNOVATION SPONSOR
> aws
> PLATINUM SPONSORS
> MongoDB.
> Google Cloud
> neo4j

![[assets/reconstructed-slides/knDDGYHnnSI/slide-002.jpg]]

- Source frame: `slide-002.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `155.66`

OCR text:

> neo4j

![[assets/reconstructed-slides/knDDGYHnnSI/slide-003.jpg]]

- Source frame: `slide-003.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `176.66`

OCR text:

> TheEvolutionof..WebSearch
> SeriousSportsFansOnly$1.000000inCashandPrizes!
> Forserioussportsfansonly!PlayFantasyFootball!
> LUOS
> It's amazing where
> GoGetItwillgetyou.
> Find:
> GoGet It
> Enhanceyoursearch.
> NETSCAFE
> INVEST
> NoNo!
> $100K
> xplorer
> NewSearch.TopNewsSitesbySubject.Top5%Sites·City GuidePictures&Sounds
> PeopleFind.PointReviewRoadMaps·Software·AboutLycos·ClubLycos·Help
> AddYourSiteto Lycos
> Copyright1996Lycosm，Inc.AllRights Reserved.
> Lycosisa trademarkofCarmegicMellonUniversity.
> Questions&Comments
> 3Neo4jInc.Allrightsreserved2024

![[assets/reconstructed-slides/knDDGYHnnSI/slide-004.jpg]]

- Source frame: `slide-004.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `158.43`

OCR text:

> TheEvolutionof..WebSearch
> Full text Era:1994-2000
> AIE
> FullText
> YAHOO!
> Netscape
> Search
> Ask
> LYCOSAOL
> erved2024
> Microsoft
> smol
> aws

![[assets/reconstructed-slides/knDDGYHnnSI/slide-005.jpg]]

- Source frame: `slide-005.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `164.48`

OCR text:

> World'sFair
> AEE

![[assets/reconstructed-slides/knDDGYHnnSI/slide-006.jpg]]

- Source frame: `slide-006.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `177.42`

OCR text:

> Google LaunchesWorld'sLargestSearchEngine
> Subscnbe
> GoogleNowEnablesInternetUserstoSearchMoreThan 1BillionURLs,ProvidingQuickandEasyAccessto560MillionFull-Text
> IndexedWebPagesand50oMillionPartiallyIndexedURLs
> AIE
> released the largestsearchengine on the Internet.Google'snewindex,comprisingmore than1billionURLs,offersusers theweb'smost
> comprehensive collection ofwebsites,whichcanbe easilysearchedwithGoogle'sfastandhighlyrelevantsearch technology.Availablenow
> atwww.google.com，Google'sportalanddestination sitecustomerscanalsolicensethisnewindexforintegrationwiththeirownwebsites.
> 沙
> Googleisbasedonavarietyofinnovative
> technologies,includingsophisticatedtextmatching
> anditsadvanced,patent-pendingtechnologycalled
> PageRanktm,whichensuresthatthemostimportant
> resultsalways comeupfirst.
> rved2024
> Microsoft
> smol"
> aws

![[assets/reconstructed-slides/knDDGYHnnSI/slide-007.jpg]]

- Source frame: `slide-007.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `159.22`

OCR text:

> TheEvolutionof..WebSearch
> PageRankEra:2000-2012
> AIE
> PageRank
> Google
> Full Text
> YAHOO!
> Netscape
> Search
> Ask
> LYCOS
> AAOL
> erved2024
> Microsoft
> smol
> aws

![[assets/reconstructed-slides/knDDGYHnnSI/slide-008.jpg]]

- Source frame: `slide-008.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `164.55`

OCR text:

> 日
> World'sFair

![[assets/reconstructed-slides/knDDGYHnnSI/slide-009.jpg]]

- Source frame: `slide-009.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `176.66`

OCR text:

> Moscone Center:Homepage
> FedEONEST-Cg2024/FDe
> AIE
> SEICON wESTCoeg2024/FigmaDesig Auom
> Events
> MosconeCenter
> FloorPlans
> 4.5662G
> Explore the Neighborhood
> Neirby Nrighooe
> Web
> 口S
> <Sae
> oodAreaAbno
> edin Pe
> Directions andParking
> BARTofrs fstcon
> nient regi
> Od MUN
> ContactUs
> The George R Mooc
> asrodod uo
> MceCel
> FisCfin
> hals p cet bckan8asinSouhfMat
> neigMocoof Wikipeda
> People also ask！
> Aod:747HS.SFCA94103
> Parcng:Pey aing ety
> Which BART stop for Moscone Center?
> OpOss11
> 60S157m
> erved2024
> neoqj
> Microsoft
> smol
> aws

![[assets/reconstructed-slides/knDDGYHnnSI/slide-010.jpg]]

- Source frame: `slide-010.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `166.87`

OCR text:

> Moscone Center
> on cerntaf in SasJFnen
> AIE
> Moscone
> Center
> Owner
> R. IMor
> lon Cenet popuary knove s the
> 05 na 9
> eeg ut xaydooo uosquts putl uo
> hw blocs n 17 ac eSouh Mara
> waipeda
> Named
> SSasf
> CA9413
> Sce bunuecie
> After
> ClyandContyf SanF
> Phone:(418976-4000
> oes11P
> BStS
> George
> Moscone
> erved2024
> neoaj
> Microsoft
> smol
> aws

![[assets/reconstructed-slides/knDDGYHnnSI/slide-011.jpg]]

- Source frame: `slide-011.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `165.14`

OCR text:

> Google1/o'24Keynote
> AIE
> Aeq p an
> pddpqpuo
> erved2024
> Microsoft
> smol°
> aws

![[assets/reconstructed-slides/knDDGYHnnSI/slide-012.jpg]]

- Source frame: `slide-012.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `158.59`

OCR text:

> WhatISGraphRAG?
> AIE
> VerySimple.
> GraphRAG isRAGwhere the Rpath includes
> aKnowledgeGraph.
> erved2024
> Microsoft
> smol
> aws

![[assets/reconstructed-slides/knDDGYHnnSI/slide-013.jpg]]

- Source frame: `slide-013.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `175.78`

OCR text:

> User
> Your Application
> Neo4j Inc.Allrightsreserved2024

![[assets/reconstructed-slides/knDDGYHnnSI/slide-014.jpg]]

- Source frame: `slide-014.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `157.77`

OCR text:

> GraphRAGRetrieval Patterns
> ThecoreGraphRAGpatternisverysimple,yetpowerful.
> Doavectorsearch tofind an initial setofnodes
> 19Neo4jInc.Allrightsreserved2024

![[assets/reconstructed-slides/knDDGYHnnSI/slide-015.jpg]]

- Source frame: `slide-015.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `161.03`

OCR text:

> GraphRAGRetrieval Patterns
> ThecoreGraphRAGpattern isverysimple,yetpowerful.
> AIE
> Doavectorsearch tofind an initial setofnodes
> Traverse the graph around those nodes to add context
> (Optional:Rank the results using the graph and pass the
> top-kdocumentstotheLLM)
> orved2024
> Microsoft
> smol
> aws

![[assets/reconstructed-slides/knDDGYHnnSI/slide-016.jpg]]

- Source frame: `slide-016.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `161.26`

OCR text:

> AIE
> TheBenefitsof
> GraphRAG
> Microsoft
> smol
> aws
> World

![[assets/reconstructed-slides/knDDGYHnnSI/slide-017.jpg]]

- Source frame: `slide-017.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `177.0`

OCR text:

> Higher
> 3x
> Knowledge Graph
> improved
> the accuracyof LLMresponses
> Accuracy
> by54.2%,anaverage of3x.
> The clearestdriverof
> JuanSequeda.1st
> Poncipw Soe
> d:co-hosf of Catx
> GraphRAG adoption amongst
> AIE
> users ishigheraccuracy.
> WOwtrulylowwyyte
> Graph+LLM Benchmark.
> seemore
> Day-1o-dny
> Analyacs
> Strxs
> ABENOIMARKTO UDERSTANDTIR ROLEOF KNOWLEDGE
> GANESONLAROELANGCACEMOOLSACCIRACYFOR
> QEESTIONANSWERINGONETERPRISESQLDATABASES
> erved2023
> Microsoft
> smol
> aws

![[assets/reconstructed-slides/knDDGYHnnSI/slide-018.jpg]]

- Source frame: `slide-018.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `177.44`

OCR text:

> ①Higher
> MicrosoftResearchBlog
> Accuracy
> GraphRAG:Unlocking LLM
> The clearestdriver of
> discovery on narrative private data
> GraphRAG adoption amongst
> ary13,2024
> users ishigher accuracy.
> "By combining LLM-generated
> knowledge graphs and graph
> machine learning, GraphRAG
> enablesustoanswerimportant
> classesofquestions thatwe cannot
> attempt with baseline RAG alone."
> Microsoft
> 23Neo4jInc.All rightsreserved2023

![[assets/reconstructed-slides/knDDGYHnnSI/slide-019.jpg]]

- Source frame: `slide-019.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `176.15`

OCR text:

> Easier
> Development
> Thesecondreasonwehear
> peoplechooseGraphRAGover
> vector-onlyRAG iseasier
> development...oncethey've
> pushed throughtheinitial
> learning curve.
> *Weirdly,easeof development[or lack
> thereofj is also one of the stumblingblocks!
> Howcan thatbe?
> Oneword:KnowledgeGraphConstruction!
> Ok that's three.Buthold thatthoughtuntil
> later!
> 24 Neo4jInc.All rightsreserved2023

![[assets/reconstructed-slides/knDDGYHnnSI/slide-020.jpg]]

- Source frame: `slide-020.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `176.25`

OCR text:

> EasierDevelopment:Why?
> Natural Language Description:"Apples and oranges are both fruits"
> DATASEMANTICS
> AIE
> EXPLICIT(SYMBOLIC)
> Representation
> erved2023
> Microsoft
> smol
> aws

![[assets/reconstructed-slides/knDDGYHnnSI/slide-021.jpg]]

- Source frame: `slide-021.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `177.7`

OCR text:

> EasierDevelopment:Why?
> Transparent&Explicit
> Opaque&Implicit
> [3.9150659e-03 2.6659777e-031.0298982e-03-2.7156321e-03
> 1.9977870e-03 3.1204436e-03 1.2055682e-04 1.0450699e-03
> -6.4308796e-043.0822519e-03 2.1972554e-03 5.1480172e-05
> AIE
> -3.7099270e-03 3.9439583e-03 6.8276987e-047.7137066e-04
> 2.3698520e-03-7.8547641e-046.0383842e-044.6370425e-03
> -1.6786088e-031.7417425e-032.4216413e-033.6545738e-03
> -1.9871239e-03 2.9489421e-03-1.2810023e-03-4.9174053e-04
> -3.9743204e-03-2.7023794e-03-3.0541950e-04-1.5724347e-03
> 4.0824865e-034.6588355e-043.5028579e-034.8283348e-03
> -2.8737928e-03-4.5569306e-03-7.6568732e-04-3.3311991e-03
> 3.5790715e-034.2424244e-033.3478225e-03-7.4140396e-04
> 1.0030111e-03-5.2394503e-045.8383477e-04-4.8430995e-03
> 2.6972082e-03-4.8002079e-03-2.3011414e-03 8.0388715e-04
> 1.7713077e-03-3.0159748e-031.7178850e-03-1.9258332e-03
> -2.4637436e-033.3779652e-03 2.7676420e-031.8853768e-03
> 2.0586823e-034.9897884e-034.5599132e-03-1.0976522e-03
> -8.7642738e-063.9748671e-03 5.2895391e-046.3330121e-04
> -1.2614765e-03-8.5018738e-043.7659388e-033.0237564e-03
> 4.5014662e-034.3258793e-03-4.2659100e-034.9081761e-03
> Customer:I actually already fixed a couple of bugs thanks to this!
> erved2023
> Microsoft
> smol
> aws

![[assets/reconstructed-slides/knDDGYHnnSI/slide-022.jpg]]

- Source frame: `slide-022.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `178.04`

OCR text:

> EasierDevelopment:Why?
> Transparent&Explicit
> Opaque&lmplicit
> [3.9150659e-03 2.6659777e-03 1.0298982e-03-2.7156321e-03
> 1.9977870e-03 3.1204436e-03 1.2055682e-04 1.0450699e-03
> -6.4308796e-043.0822519e-03 2.1972554e-035.1480172e-05
> -3.7099270e-03 3.9439583e-036.8276987e-047.7137066e-04
> 2.3698520e-03-7.8547641e-046.0383842e-044.6370425e-03
> -1.6786088e-031.7417425e-03 2.4216413e-03 3.6545738e-03
> -1.9871239e-032.9489421e-03-1.2810023e-03-4.9174053e-04
> -3.9743204e-03-2.7023794e-03-3.0541950e-04-1.5724347e-03
> -4.0824865e-034.6588355e-043.5028579e-034.8283348e-03
> -2.8737928e-03-4.5569306e-03-7.6568732e-04-3.3311991e-03
> 3.5790715e-034.2424244e-033.3478225e-03-7.4140396e-04
> 1.0030111e-03-5.2394503e-04 5.8383477e-04-4.8430995e-03
> 2.6972082e-03-4.8002079e-03-2.3011414e-03 8.0388715e-04
> 3.1952575e-05-8.1621204e-04-3.8127291e-03-6.7428290e-04
> -1.7713077e-03-3.0159748e-031.7178850e-03-1.9258332e-03
> -2.4637436e-03 3.3779652e-032.7676420e-03 1.8853768e-03
> -2.4718521e-03-1.9754141e-03 2.6104036e-03-2.1335895e-03
> 2.4405334e-03-3.2013952e-043.9961869e-034.0419102e-03
> 2.0586823e-03 4.9897884e-03 4.5599132e-03-1.0976522e-03
> 1.5563263e-03 3.9063310e-03-2.9308300e-03-4.8254002e-03
> -8.7642738e-063.9748671e-035.2895391e-046.3330121e-04
> -1.2614765e-03-8.5018738e-043.7659388e-03 3.0237564e-03
> 4.5014662e-034.3258793e-03-4.2659100e-03 4.9081761e-03
> -3.9214552e-03-2.4262110e-03-8.1192164e-05-4.1112076e-03]
> Customer:"Iactuallyalreadyfixedacoupleofbugsthankstothis!
> 28Neo4jInc.Allrightsreserved2023

![[assets/reconstructed-slides/knDDGYHnnSI/slide-023.jpg]]

- Source frame: `slide-023.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `159.71`

OCR text:

> KnowledgeGraphConstruction
> Unstructured Data
> MixofStructured&
> Structured Data
> TypicallyPDFsorother
> UnstructuredData
> Structureddatawith
> textdocuments
> Structureddatawith
> AIE
> shorttextvalues
> long-formtext
> erved2024
> Microsoft
> smol
> aws

![[assets/reconstructed-slides/knDDGYHnnSI/slide-024.jpg]]

- Source frame: `slide-024.jpg`
- Crop: `full` `[0, 0, 960, 540]` score `166.22`

OCR text:

> Two Types of Knowledge Graphs
> LexicalGraph
> Agraphrepresentationoffor
> AIE
> examplethewords,paragraphs,
> chunks,documentsand the
> relationshipsbetweenthem
> erved2024
> Microsoft
> smol°
> aws
> World's
## Dense Scene-Detected Slide Candidates
- [[youtube-knDDGYHnnSI-dense-slides]]