<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Carly Jazwin on Medium]]></title>
        <description><![CDATA[Stories by Carly Jazwin on Medium]]></description>
        <link>https://medium.com/@carly.jazwin_9433?source=rss-98cc36dfee80------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*plMQynhWTlapkJck9rt50w.jpeg</url>
            <title>Stories by Carly Jazwin on Medium</title>
            <link>https://medium.com/@carly.jazwin_9433?source=rss-98cc36dfee80------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 10:47:47 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@carly.jazwin_9433/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Extracting Structured Data with KVP Schemas in watsonx Orchestrate]]></title>
            <link>https://medium.com/@carly.jazwin_9433/extracting-structured-data-with-kvp-schemas-in-watsonx-orchestrate-594eb2b74784?source=rss-98cc36dfee80------2</link>
            <guid isPermaLink="false">https://medium.com/p/594eb2b74784</guid>
            <category><![CDATA[ibm]]></category>
            <category><![CDATA[document-processing]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ibm-watson]]></category>
            <category><![CDATA[automation]]></category>
            <dc:creator><![CDATA[Carly Jazwin]]></dc:creator>
            <pubDate>Mon, 05 Oct 2026 14:57:32 GMT</pubDate>
            <atom:updated>2026-10-05T14:57:32.125Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wJnOaBnKcwyks9Q9jBBf4w.png" /></figure><blockquote>The series so far:<strong> </strong><a href="https://medium.com/@noah.eigenfeld/your-first-agentic-workflow-with-watsonx-orchestrate-609e19b158eb"><strong>Part 1: Creating an agent and workflow in WxO</strong></a><strong>; </strong><a href="https://medium.com/@noah.eigenfeld/sorting-documents-with-watsonx-orchestrate-fe5eab3f72b6"><strong>Part 2: Sorting documents in WxO</strong></a><strong>; </strong><a href="https://medium.com/@noah.eigenfeld/extracting-data-from-documents-in-watsonx-orchestrate-7ee1ceaa54a2"><strong>Part 3: Extracting data from documents in WxO;</strong></a><strong> </strong><a href="https://medium.com/@acywang27/extracting-data-from-unstructured-documents-in-watsonx-orchestrate-51d43981b5af"><strong>Part 4: Extracting Data from Unstructured Documents in watsonx Orchestrate</strong></a><strong>, </strong><a href="https://medium.com/@carly.jazwin_9433/extracting-text-from-documents-in-watsonx-orchestrate-722bee639382"><strong>Part 5: Extracting raw text from documents in WxO</strong></a><strong>; Part 6: Extracting structured data with Text Extractor KVP Schemas (this article)</strong></blockquote><p>In the previous article, we used the Text Extractor to convert documents into text and summarize the results with an AI Task.</p><p>While that approach is useful when you need the document contents themselves, many workflows require specific pieces of information such as IDs, names, dates, statuses, or other business data.</p><p>For these scenarios, the Text Extractor provides <strong>semantic key-value pair (KVP) extraction</strong>. KVP extraction uses an AI model and a schema definition to identify specific pieces of information within unstructured documents. The schema defines which fields should be extracted and how they should be returned in the output.</p><p>In this article, we’ll configure a KVP schema, use it to extract structured information from a document, and return those values as workflow outputs.</p><p><strong>Step 1: Create and Configure a Text Extractor</strong></p><p>Add a <strong>Text Extractor</strong> node to a new workflow.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bz7OV3rbwSUvy3OQg9-5EA.png" /></figure><p>For this example, enable <strong>Output as Object</strong> so the extracted values can be used throughout the workflow.</p><p>Then enable <strong>Extract Key-Value Pairs</strong> in the Text Extractor properties panel.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*58JG-sZ99xmMZQMumHqVBw.png" /></figure><p>Once enabled, an <strong>Add schema</strong> button will appear.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*9G9ckqXARWIJeICRENmkQw.png" /></figure><p>Click <strong>Add schema</strong> to open the <strong>KVP Schema Editor</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9rZlUOJsoKNMz02I3693wA.png" /></figure><p><strong>Step 2: Create a KVP Schema</strong></p><p>The KVP Schema Editor allows you to define the information that should be extracted from a document.</p><p>Unlike the Document Extractor, which uses a visual field-definition experience, KVP extraction uses a JSON schema to describe the information you want to extract.</p><p>For this example, we’ll extract information from a project status report.</p><p>Create a schema containing the following fields:</p><ul><li>Project ID</li><li>Project Manager</li><li>Overall Status</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qKGsKr4Sm4lVuF-_R6ZvTA.png" /></figure><p>Provide descriptions for each field so the model understands what information should be extracted. Sample values can also improve extraction accuracy.</p><p>As you build your schema, the editor validates your input and provides suggestions to help guide the schema definition process.</p><p><strong>Step 3: Select a Model</strong></p><p>Before saving the schema, select a model to perform the extraction. <strong>A schema cannot be saved until a model has been selected.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/982/1*-MoiXvn8oOpppWXt0Puy9A.png" /></figure><p>For guidance on choosing a model, see the documentation <a href="https://www.ibm.com/docs/en/watsonx/watson-orchestrate/base?topic=agents-choosing-ai-model-agent">here</a>.</p><p><strong>Step 4: Configure Workflow Inputs and Outputs</strong></p><p>First, add a <strong>File</strong> variable named document to the Start node.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4Unz6-GkSLJZ2-08sqERog.png" /></figure><p>Next, configure the outputs that should be returned when the workflow completes. For this example, we’ll add outputs for the project ID, project manager, and overall status.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6lMS-0__zDksowRSxHeF5g.png" /></figure><p>After creating the outputs, verify that each one is connected to the corresponding field returned by the Text Extractor. If necessary, click <strong>Edit data mapping</strong> to adjust the mappings.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MvyfpbQOBi0Y7mETsIcksA.png" /></figure><p><strong>Step 5: Test the Workflow</strong></p><p>Click <strong>Preview</strong> and open the workflow in Agent Builder.</p><p>Ask the agent to process a document, then upload a document that contains the information described by your schema.</p><p>When the workflow completes, it should return structured values for each output field.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LiiFg95ZGfRF27Ti1hUtmA.png" /></figure><p>The exact values will depend on the contents of the document being processed.</p><p><strong>Working with Multiple Schemas</strong></p><p>A Text Extractor can contain multiple schemas.</p><p>For example, you might create:</p><ul><li>A schema for project status reports</li><li>A schema for meeting summaries</li><li>A schema for contracts</li></ul><p>By default, the Text Extractor determines which schema best matches the document being processed.</p><p>If you always want a specific schema to be used, enter its name in the <strong>KVP Force Schema Name</strong> field. This bypasses automatic schema selection and forces the specified schema to run.</p><p><strong>Next Steps</strong></p><p>We now know how to use Text Extractor KVP schemas to extract structured information from unstructured documents.</p><p>By defining a schema and selecting an extraction model, you can return targeted business data such as IDs, names, statuses, dates, and other key fields as workflow outputs.</p><p>This approach makes it possible to work with document content in a structured format that can be used throughout your workflows and automations.</p><p>Follow to read new posts in this series as soon as they are published.</p><p><strong>Links and resources</strong></p><ul><li><a href="https://www.ibm.com/products/watsonx-orchestrate">watsonx Orchestrate Product Page</a></li><li><a href="https://www.ibm.com/docs/en/watsonx/watson-orchestrate/base?topic=builder-adding-text-extractors">watsonx Orchestrate Official Documentation — Text Extractors</a></li><li><a href="https://www.ibm.com/docs/en/watsonx/watson-orchestrate/base?topic=agents-choosing-ai-model-agent">Official Documentation on Choosing an AI Model</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=594eb2b74784" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Extracting Text from Documents in watsonx Orchestrate]]></title>
            <link>https://medium.com/@carly.jazwin_9433/extracting-text-from-documents-in-watsonx-orchestrate-722bee639382?source=rss-98cc36dfee80------2</link>
            <guid isPermaLink="false">https://medium.com/p/722bee639382</guid>
            <category><![CDATA[document-processing]]></category>
            <category><![CDATA[ibm]]></category>
            <category><![CDATA[watsonx-orchestrate]]></category>
            <category><![CDATA[ibm-watson]]></category>
            <category><![CDATA[ai-agent]]></category>
            <dc:creator><![CDATA[Carly Jazwin]]></dc:creator>
            <pubDate>Fri, 02 Oct 2026 18:50:16 GMT</pubDate>
            <atom:updated>2026-10-02T18:50:16.715Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QFCo8MsuQcTRzmvT3sZcfg.png" /></figure><blockquote>The series so far:<strong> </strong><a href="https://medium.com/@noah.eigenfeld/your-first-agentic-workflow-with-watsonx-orchestrate-609e19b158eb"><strong>Part 1: Creating an agent and workflow in WxO</strong></a><strong>; </strong><a href="https://medium.com/@noah.eigenfeld/sorting-documents-with-watsonx-orchestrate-fe5eab3f72b6"><strong>Part 2: Sorting documents in WxO</strong></a><strong>; </strong><a href="https://medium.com/@noah.eigenfeld/extracting-data-from-documents-in-watsonx-orchestrate-7ee1ceaa54a2"><strong>Part 3: Extracting data from documents in WxO;</strong></a><strong> </strong><a href="https://medium.com/@acywang27/extracting-data-from-unstructured-documents-in-watsonx-orchestrate-51d43981b5af"><strong>Part 4: Extracting Data from Unstructured Documents in watsonx Orchestrate</strong></a><strong>, Part 5: Extracting raw text from documents in WxO (this article)</strong></blockquote><p>In the previous articles in this series, we focused on using the Document Extractor to extract specific information from documents such as invoices, contracts, and forms.</p><p>However, sometimes you need the document contents themselves. For those scenarios, watsonx Orchestrate provides the <strong>Text Extractor</strong> node.</p><p>The Text Extractor converts documents into text that can be summarized, classified, searched, or passed to downstream workflow steps.</p><p>In this article, we’ll build a simple Text Extractor workflow and use the extracted content in an AI Task node to generate a summary. You’ll learn how to extract document text, pass it to downstream workflow steps, and test the workflow in an agent.</p><p><strong>Step 1: Create the Text Extractor</strong></p><p>Starting with a new workflow, click <strong>Add your first step</strong> and select the <strong>Text Extractor</strong> node from the list of flow activities.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LtuuUVG-0JHzreqxjFp6dw.png" /></figure><p>Unlike the Document Extractor, which opens a dedicated configuration experience, the Text Extractor is configured directly from the properties panel on the right side of the Flow Builder.</p><p>After adding the node, select it to open its configuration options.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BCh27XlGSQw1HXEugR8FLA.png" /></figure><p><strong>Step 2: Configure the Text Extractor</strong></p><p>The Text Extractor includes several options that control how documents are processed.</p><p>For this example, we’ll make a few key configuration changes.</p><p><strong>Document Structure Mode</strong></p><p>Set <strong>Document Structure Mode</strong> to <strong>Keep Original Formatting</strong>.</p><p>This preserves document structure such as headings, lists, and tables in the extracted output. Maintaining the original document structure can improve readability and provide richer context for downstream processing.</p><p><strong>Output as Object</strong></p><p>Enable <strong>Output as Object</strong> so the extracted results can be accessed and mapped directly within the workflow. This makes it easy to pass the extracted text to downstream nodes such as an AI Task.</p><p><strong>Extracted Content Types</strong></p><p>The Text Extractor can return content as Text, Markdown, or HTML.</p><p>For this example, we’ll use <strong>Markdown. </strong>Markdown preserves document structure such as headings, lists, and tables while remaining easy for downstream AI models to process, as shown in the following example.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*j0ANosn9hrRH2plMFrFdvA.png" /></figure><p>At this point, the Text Extractor is fully configured and can be used to return document contents in the selected output format.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*v88CL86gwgQgWal9tWjxpw.png" /></figure><p>The remaining steps show one example of how the extracted text can be used in a downstream workflow.</p><p><strong>Step 3: Summarize the Extracted Text (Example Workflow)</strong></p><p>One of the most common ways to use extracted text is to pass it to a downstream node for further processing. For example, you might summarize a report, classify a document, extract key topics, or answer questions based on its contents.</p><p>For this example, we’ll send the extracted Markdown content to an <strong>AI Task</strong> node and use it to generate a summary.</p><p>Add an <strong>AI Task </strong>node after the Text Extractor.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pPlrkmFbMHGSy2l28swXPQ.png" /></figure><p>Create an input variable named document_text and configure a prompt that uses the extracted document contents to generate a summary.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0DgN4_yV4SD6lF3tE6NLPQ.png" /></figure><p>Select a model and save the node. Next, we’ll configure the workflow inputs and outputs before testing it.</p><p><strong>Step 4: Set Flow Inputs and Outputs</strong></p><p>First, add a <strong>File</strong> variable named document to the Start node.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MYxL-jy_TcXBVWpJ8LiQ8w.png" /></figure><p>This should automatically map to the Text Extractor input.</p><p>Next, configure the outputs that should be returned when the workflow completes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*u_uFRlTX8DhK5aPEWId87w.png" /></figure><p>Add an output variable named summary to the End node. This output will contain the response generated by the AI Task.</p><p>In many cases, watsonx Orchestrate automatically maps outputs to the appropriate values. However, if any mappings are incorrect, click <strong>Edit data mapping</strong> and manually select the appropriate value.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YgD0V9wL920gngLcYCXdKQ.png" /></figure><p><strong>Preview the Workflow</strong></p><p>The workflow is now ready to test.</p><p>From the Flow Builder screen, click <strong>Preview</strong> in the top-right corner, then select <strong>Go to agent builder</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sGjqVKaPbNntoeR7rlkmcA.png" /></figure><p>Ask the agent to process a document, and when prompted, upload a document that is appropriate for your workflow.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8arhlFP5cWG5bqT_mmxlVw.png" /></figure><p>If everything is configured correctly, the workflow will use the extracted document text to generate a summary highlighting the most important information.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0pw7pZh6GS8X2qcDBpH1qA.png" /></figure><p>At this point, you’ve built a complete document-processing workflow. The Text Extractor converts the document into usable text, and the AI Task generates a summary that is returned to the user.</p><p><strong>Next Steps</strong></p><p>We now know how to use the Text Extractor node to convert documents into usable text and pass that content to downstream workflow steps.</p><p>Whether you’re generating summaries, classifying documents, answering questions, or preparing content for AI-powered workflows, the Text Extractor provides a flexible foundation for document processing in watsonx Orchestrate.</p><p>In the next article, we’ll move beyond document text and explore using Text Extractor KVP schemas to extract specific structured values from documents.</p><p>Follow to read new posts in this series as soon as they are published.</p><p><strong>Links and resources</strong></p><ul><li><a href="https://www.ibm.com/products/watsonx-orchestrate">watsonx Orchestrate Product Page</a></li><li><a href="https://www.ibm.com/docs/en/watsonx/watson-orchestrate/base?topic=builder-adding-text-extractors">watsonx Orchestrate Official Documentation — Text Extractors</a></li><li><a href="https://www.ibm.com/docs/en/watsonx/watson-orchestrate/base?topic=agents-choosing-ai-model-agent">Official Documentation on Choosing an AI Model</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=722bee639382" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>