{
  "docId": "019dd923-5e88-73ef-bd59-174b7f7da09c",
  "docSlug": "d9ee2fdc0bd7b0ec",
  "documentTitle": "enhaced data extraction using gen ai ey collaboration with wlastic",
  "authorId": "MorganStanley",
  "authorName": "EY",
  "documentKindSlug": "consulting-deck",
  "documentKindLabel": "Consulting deck",
  "sourceTypeSlug": "equity_research",
  "sourceTypeLabel": "Equity research",
  "presentationDate": null,
  "orientation": "landscape",
  "aspectRatio": 1.778,
  "pageNumber": 6,
  "pageCount": 16,
  "prevPage": 5,
  "nextPage": 7,
  "slideType": "other",
  "function": "present_framework",
  "density": "overcrowded",
  "nDataPoints": 0,
  "notes": "The slide explains the technical components of a RAG (Retrieval-Augmented Generation) pipeline.",
  "elementsJson": [
    "headline_text",
    "paragraph",
    "footnote",
    "photo"
  ],
  "metadataConfidence": 0.9,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd59-174b7f7da09c/6",
  "deckHref": "/decks/019dd923-5e88-73ef-bd59-174b7f7da09c",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd59-174b7f7da09c.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd59-174b7f7da09c#slide-6",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "These models help convert normal language into a space of vectors that are understood by retriever systems, emphasizing the understanding of context and the user's intent, and transcending the limitations of traditional keyword matching.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-340f-723f-8a13-296177de2de9",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 1,
        "w": 0.32,
        "x": 0.68,
        "y": 0
      },
      "kind": "image",
      "text": null,
      "attrs": null,
      "subkind": "photo",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "89ab55de-7648-4071-b22f-bc0796b73ef7",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.2,
        "w": 0.3,
        "x": 0.05,
        "y": 0.32
      },
      "kind": "paragraph",
      "text": "A crucial component of the retrieval pipeline is the vector store, which is essentially a robust data storage system that can handle a diverse array of data types. This includes unstructured text, structured data, and dense vectors (embeddings). The vector store is designed to accommodate data both before and after it has been transformed by embedding models, making it a versatile tool in the pipeline.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "1c759217-74a1-4eb1-9051-268abcef4734",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.03,
        "w": 0.35,
        "x": 0.05,
        "y": 0.16
      },
      "kind": "paragraph",
      "text": "The end-to-end pipeline of the retrieval system is outlined below:",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "560f79fd-082d-4ec7-aa2d-e290d24c66f8",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.35,
        "w": 0.3,
        "x": 0.37,
        "y": 0.32
      },
      "kind": "paragraph",
      "text": "These retrieval systems often employ embedding models, such as the Elastic Learned Sparse EncodeR (ELSER)², to facilitate a retrieval model that offers enterprises the capability to execute precise semantic searches. These models help convert normal language into a space of vectors that are understood by retriever systems, emphasizing the understanding of context and the user's intent, and transcending the limitations of traditional keyword matching. Harnessing a rich training dataset composed of high-quality question-answer pairs, these models enhance the efficiency of calculating similarity between queries and documents. This not only refines the accuracy of information retrieval, but also expedites the indexing process, improving the search experience for users",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "e46a6d2e-5f19-471b-86d2-3e19dc03de0e",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.2,
        "w": 0.3,
        "x": 0.68,
        "y": 0.72
      },
      "kind": "source-note",
      "text": "2 ELSER is a retrieval model trained by Elastic that enables semantic search, which allows the solution to retrieve more relevant search results. This search type provides search results based on contextual meaning and user intent, rather than exact keyword matches. 3 RRF is a method for combining multiple result sets with different relevance indicators into a single result set. RRF requires no tuning, and the different relevance indicators do not have to be related to each other to achieve high-quality results.",
      "attrs": null,
      "subkind": null,
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "96b938f4-f4f7-4ac7-8563-fdb5323e494c",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.03,
        "w": 0.15,
        "x": 0.05,
        "y": 0.24
      },
      "kind": "title",
      "text": "Vector store",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "0a03f1da-505c-4daa-9b19-e24f3275f9f0",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.03,
        "w": 0.2,
        "x": 0.37,
        "y": 0.24
      },
      "kind": "title",
      "text": "Embedding models",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "b0dc2987-60b8-4bd0-9217-6d337d9f29eb",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.4,
        "x": 0.05,
        "y": 0.08
      },
      "kind": "title",
      "text": "Gen AI and retrieval strategies (Cont'd)",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "f05cfadb-1176-4782-b31f-c731fe43faa4",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [],
  "frameworks": [],
  "arcBeats": [
    {
      "to": 6,
      "from": 5,
      "beatId": "1efec661-4651-4613-962f-1c8e8166349a",
      "arcName": "The Consultant's Gambit",
      "arcSlug": "consultants-gambit",
      "beatName": "Solution & Approach",
      "beatSlug": "consultants-gambit-solution-approach",
      "evidence": "The deck presents Gen AI as a solution to improve data extraction efficiency",
      "position": 2,
      "confidence": 0.8,
      "parentBeatName": "Turn",
      "parentBeatSlug": "turn"
    }
  ],
  "loops": [
    {
      "to": 6,
      "from": 3,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "565e85cb-f579-4e65-a33a-b008d293d5f2",
      "evidence": "The deck presents a problem statement and a proposed solution using Gen AI",
      "position": 0,
      "objective": "Identify patterns in current data extraction methods",
      "structure": "Evidence A -> Evidence B -> Evidence C -> Pattern/Conclusion",
      "confidence": 0.6,
      "description": "Group multiple pieces of evidence that together point to a pattern or conclusion"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}