<?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 Subigya Upadhyay on Medium]]></title>
        <description><![CDATA[Stories by Subigya Upadhyay on Medium]]></description>
        <link>https://medium.com/@subigyaraj?source=rss-bc9cd0fc3aa5------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>Stories by Subigya Upadhyay on Medium</title>
            <link>https://medium.com/@subigyaraj?source=rss-bc9cd0fc3aa5------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 18:40:42 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@subigyaraj/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[You Can’t Build an Enterprise Knowledge Graph. You Can Grow One.]]></title>
            <link>https://medium.com/@subigyaraj/you-cant-build-an-enterprise-knowledge-graph-you-can-grow-one-68c514251639?source=rss-bc9cd0fc3aa5------2</link>
            <guid isPermaLink="false">https://medium.com/p/68c514251639</guid>
            <category><![CDATA[graphrag]]></category>
            <category><![CDATA[enterprise-ai]]></category>
            <category><![CDATA[ontology]]></category>
            <category><![CDATA[knowledge-graph]]></category>
            <dc:creator><![CDATA[Subigya Upadhyay]]></dc:creator>
            <pubDate>Wed, 18 Nov 2026 13:12:23 GMT</pubDate>
            <atom:updated>2026-11-18T13:20:36.673Z</atom:updated>
            <content:encoded><![CDATA[<p><em>A story about a stalled programme, twenty customer tables, and the question nobody asked first: do we even need one?</em></p><h3>The ask</h3><p>It started, as these things often do, after a conference. The chief data officer of a mid-sized manufacturer came back convinced: AI agents were coming, they would need context, and the company needed an enterprise knowledge graph.</p><p>The brief landed on the lead data architect. It was one line long: “Build us an enterprise knowledge graph.”</p><p>The architect asked the only question that mattered: “For what?” Nobody had an answer. That turned out to be the whole story.</p><p>The story is a composite, but if you’ve worked on a knowledge graph programme, you’ll recognise every scene.</p><h3>Six months of modelling everything</h3><p>The team went top-down. If this was going to be <em>the</em> enterprise knowledge graph, it had to model the enterprise: every product, party, contract, asset and event, defined properly and independent of any database.</p><p>It took six months, and the result was elegant. Then they went looking for data to populate it. Half the classes had nothing behind them. Some of the data existed but meant something subtly different from what the model said.</p><p>So they reverse-engineered from the tables, bottom-up. Then the bottom-up model didn’t fit the concepts, so they went back up again. Up and down, down and up.</p><p>By month nine, the programme had a beautiful diagram and no users.</p><p>The lesson: you can’t design a model for questions nobody has asked. “Build for everything” usually means nobody has decided what it’s for.</p><h3>Twenty customer tables</h3><p>The team restarted from the other end: the data. The first discovery was that the company didn’t have a customer table. It had twenty.</p><p>The obvious move was to collapse them into one Customer class. The workshop to agree on that took three sessions and nearly ended in a row.</p><p>Some tables really were copies, left over from an acquisition and two migrations. But others were genuinely different things the business happened to call “customer”: prospects, account holders, billing entities, legal entities, distributors.</p><p>Collapsing those wouldn’t have resolved the ambiguity. It would have hidden it inside the ontology, where it’s harder to see. What held up was a Party that plays a customer role in the context of an agreement.</p><p>The lesson: the agreement is the deliverable. The record of what’s the same, what’s different and why is worth more than the class diagram.</p><h3>The team that didn’t need a graph</h3><p>The first team to show real interest was marketing. Their problem: product categories meant different things on the website, in the ERP and in the sales reports, so nothing reconciled.</p><p>The architect started sketching an ontology. Then a colleague asked the uncomfortable question: what would a graph give marketing that an agreed list of categories wouldn’t?</p><p>Nobody could answer. Marketing got a controlled vocabulary, governed by one owner, in six weeks. The gains from going further just weren’t there.</p><p>That wasn’t a failure. It was the most useful thing the programme had learned so far: representing meaning is a ladder, and you climb only as far as the use case needs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yDOLTecC0saQGntCwQ1D5A.png" /><figcaption><em>Figure 1. The semantic ladder: each rung adds meaning and cost. Diagram created with AI assistance.</em></figcaption></figure><p>Each rung adds meaning and cost. Stop at the first one that answers your questions.</p><p>Controlled vocabularies are also the cheapest layer to share. Agreed codes pay off across the enterprise even where no graph is ever built. And when a graph does arrive, its ontology can point at those vocabularies, using a standard such as SKOS, rather than reinvent them.</p><h3>The first question worth a graph</h3><p>The real use case arrived with a crisis. A key supplier went into administration, and the CEO asked a simple question: which of our customers are exposed, and through which contracts?</p><p>It took a team two weeks to answer. Customers lived in the CRM, contracts in the ERP and suppliers in procurement. Answering meant hopping from supplier to contract to product to customer, across all three.</p><p>That is a graph-shaped question. It crosses systems, goes several hops deep and needs an answer you can explain.</p><p>The architect wrote it down, with three related questions. Those four became the competency questions: the ontology would be done when it could answer them over real data, and not before.</p><p>This time the team copied nothing. They built a thin core of a dozen concepts, such as Party, Product, Agreement and Location, and mapped them onto the existing tables. The result was a <em>virtual</em> graph: the data stayed where it was.</p><p>The query is written against the ontology, not the tables:</p><pre>SELECT ?customer ?contract WHERE {<br>  ?supply    :withSupplier  :SupplierX ;<br>             :coversProduct ?product .<br>  ?contract  :coversProduct ?product ;<br>             :withCustomer  ?customer .<br>}</pre><p>Nothing in it names a table. Behind the ontology sit mappings that say, for example, that a customer is a row in the CRM’s account table and that coversProduct is a join to the ERP&#39;s contract lines. A virtual graph engine uses those mappings to rewrite the query into SQL, runs it against the source systems and hands back the answer as a graph.</p><p>The graph is a lens over the data, not a copy of it. Where the ontology had a concept with no table behind it, it simply showed as unmapped: a gap to report, not a reason to redesign.</p><p>Structured data suits this because it already has keys, schemas and owners. It is federated by nature. The two-week answer now took seconds.</p><p>The trade-off showed up later. Deep traversals across three systems were slow, so the team copied one hot path into a graph store for the risk dashboard. Virtual by default, materialized where the use case demands it.</p><h3>The contracts were PDFs</h3><p>Then legal asked the obvious follow-up: which of those contracts let us terminate if the supplier fails? That answer wasn’t in any table. It sat in force majeure and termination clauses, inside 4,000 PDFs.</p><p>There is nothing to federate in a PDF. You can’t map a clause to a column, so the knowledge has to be extracted and stored. Unstructured data has to be <em>materialized</em>.</p><p>Extraction is where it got hard. An LLM pulled out entities and relationships quickly, but it produced mentions, not entities. “Supplier X”, “Supplier X Holdings Ltd” and “SXH” came out as three different suppliers.</p><p>It also invented a new relationship type for almost every document. And now and then, it stated a relationship that the text only implied.</p><p>What fixed it was the structured side. Extraction was constrained to the ontology’s classes and relationships, so the model couldn’t invent its own schema. Every extracted supplier and customer had to resolve to a known record in the virtual graph, or go to a review queue.</p><p>Every fact kept a confidence score and a link to the exact clause it came from. Validation rules (SHACL) checked the structure. A hand-labelled sample of 200 contracts measured quality before anyone trusted the results.</p><p>The lesson: structured data gives you identities; unstructured data gives you claims. Validate the claims against the identities.</p><p>The architecture they ended up with looks like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DUW1YxTY-3cmDk8TzAsifw.png" /><figcaption><em>Figure 2. One ontology, two ways to fill it. Diagram created with AI assistance.</em></figcaption></figure><p>One ontology, two very different ways of filling it, joined at the point where extracted entities resolve to known records.</p><h3>The GraphRAG demo</h3><p>Halfway through, a vendor demo nearly derailed the programme. An LLM had built a graph from the same 4,000 contracts over a weekend and answered questions about them in plain English. The room was impressed.</p><p>It also had every problem the team had just spent months fixing: duplicate suppliers, invented relationships and no link to the customer records. The answers were fluent and sometimes wrong, with no way to tell which.</p><p>GraphRAG, using a graph to retrieve context for an LLM, comes in two flavours. One builds the graph from documents with an LLM. That helps with broad questions like “what themes run across these contracts?”, but it is only as good as its extraction.</p><p>The other grounds the LLM in a graph you’ve already validated. The team went with the second: the LLM writes queries against the ontology, and every answer carries its provenance.</p><p>For “find the clause that answers this” over one set of documents, plain vector search was still the right tool. GraphRAG earned its place on multi-hop questions and on answers that had to be explained.</p><p>The lesson: an LLM-built graph without an ontology is a pile of triples. It demos well and degrades quickly.</p><h3>The second use case</h3><p>Six months later, finance needed to trace which legal entities sat behind each customer for credit risk. It needed Party, Agreement and the customer roles. All three already existed.</p><p>That’s when the thin core paid off. The team’s rule was simple: a concept stays local the first time a use case needs it. When a second one needs it, it gets agreed properly and promoted into the shared core. Party had just been promoted.</p><p>The finance use case took a third of the time of the supplier one. That’s the metric that matters. If each new use case is cheaper than the last, the graph is becoming an enterprise asset. If it isn’t, you’re running isolated projects with extra steps.</p><p>If the team were starting today, the tooling would help. AWS’s open-source <a href="https://aws.amazon.com/about-aws/whats-new/2026/07/aws-context--ontology-accelarator-generally-available/">Context Ontology Accelerator</a>, released on 31 July 2026, builds much of this pattern in:</p><ul><li><strong>Bottom-up drafting.</strong> It <a href="https://github.com/aws/context-ontology-accelerator/blob/main/external-docs/content/ontologies.md">proposes an ontology</a> from your table structures and metadata.</li><li><strong>Top-down grounding.</strong> New concepts are aligned to reference ontologies such as FIBO, or to ones you’ve already accepted, so related concepts converge instead of multiplying.</li><li><strong>Meaning, not just names.</strong> It checks whether a match really means the same thing: a Dublin Core “Policy” is not an insurance “policy”.</li><li><strong>Visible gaps.</strong> Classes backed by real data are marked as mapped; the rest are not.</li><li><strong>Human approval.</strong> Domain experts review and approve every element.</li></ul><p>What it can’t do is choose your use case, write your questions or run the agreement conversation for you. Tech Mahindra’s <a href="https://www.techmahindra.com/services/data-analytics/semara-ontology-knowledge-graph-accelerator/">SEMARA</a> and Acuvate’s <a href="https://acuvate.com/accelerators/acucortex-ontology-accelerator/">AcuCortex</a> follow the same shape: start with one use case, then grow.</p><h3>What I’d tell the CDO now</h3><p>If I could go back to that first meeting, this is what I’d say.</p><ol><li><strong>Ask whether you need a graph at all.</strong> Write the questions down first. If the value is in consistent labels, stop at a vocabulary.</li><li><strong>Start with one use case that has an owner and real questions.</strong> Those questions are your acceptance tests.</li><li><strong>Treat structured and unstructured data as different problems.</strong> They need different architectures and fail in different ways.</li><li><strong>Go virtual by default for structured data.</strong> Materialize unstructured data, and structured hot paths only when speed demands it.</li><li><strong>Resolve every extracted entity against your master data.</strong> Without that, a graph built from documents turns into noise.</li><li><strong>Keep a thin shared core, and promote concepts on second reuse.</strong> That’s what stops use case eight reinventing use case one.</li><li><strong>Measure the cost of each new use case.</strong> It should keep falling.</li></ol><p>Where to start modelling depends on the use case too:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Akm4Vj0uLGRhTMANafwRAA.png" /><figcaption><em>Figure 3. Where to begin depends on the use case. Diagram created with AI assistance.</em></figcaption></figure><p>And the pitfalls the team hit, roughly in order: modelling everything first; letting the ontology mirror the source schema; choosing technology before the use case; materializing by default; underestimating entity resolution; trusting an LLM-built graph with no ontology or evaluation; and treating the graph as a project, with nobody owning the meaning.</p><h3>One view: use cases, what they demand and how the industry does it</h3><p>Across the big platforms and practitioner case studies, the same pattern holds: match the approach to what the use case demands, and most use cases never need a graph. The list below the matrix draws on vendor documentation and announcements, case studies from AstraZeneca, Netflix, S&amp;P Global and ABB, and published talk abstracts. Status as of 18 November 2026; several of these features are still in preview.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8_35kxgSIj5LcJ2EyBhGtQ.png" /><figcaption><em>Figure 4. Use cases, what they demand and how they work. Diagram created with AI assistance.</em></figcaption></figure><p>How the big platforms and practitioners handle each use case:</p><ul><li><strong>Consistent labels and reference data.</strong> AstraZeneca, S&amp;P Global and Netflix all started with governed vocabularies, and AstraZeneca writes vocabulary IDs into its data during transformation (<a href="https://juansequeda.substack.com/p/trip-report-connected-data-london-2025">Juan Sequeda’s write-up</a>). Databricks is adding a <a href="https://atlan.com/know/ai-agent/databricks/databricks-data-ai-summit-2026-announcements/">Business Glossary</a> to Unity Catalog.</li><li><strong>Q&amp;A over documents.</strong> <a href="https://docs.snowflake.com/en/en/user-guide/snowflake-cortex/cortex-agents">Snowflake Cortex Search</a> combines vector and keyword search. Databricks uses <a href="https://answers.databricks.com/architecture-giving-ai-context-across-company-data-documents">Agent Bricks Knowledge Assistant and AI Search</a>, with citations.</li><li><strong>Metrics and KPIs.</strong> Snowflake’s <a href="https://upsolve.ai/blog/snowflake-semantic-layer">semantic views</a> feed Cortex Agents, which Snowflake <a href="https://docs.snowflake.com/en/release-notes/2026/other/2026-08-28-cortex-analyst-transition-cortex-agents.md">now recommends</a> over standalone Cortex Analyst. Databricks defines <a href="https://www.databricks.com/blog/unity-catalog-metric-views-now-power-bi-one-unified-governed-semantic-layer-all-your-reporting">metric views</a> once in Unity Catalog for Genie, dashboards and Power BI.</li><li><strong>Agents reasoning over entities and rules.</strong> Microsoft’s <a href="https://www.devoteam.com/expert-view/microsoft-fabric-iq/">Fabric IQ</a> puts an ontology between OneLake data and its agents, with an MCP server. AWS’s <a href="https://aws.amazon.com/about-aws/whats-new/2026/07/aws-context--ontology-accelarator-generally-available/">Context Ontology Accelerator</a> drafts an ontology, has experts approve it and serves it to agents over MCP.</li><li><strong>Customer 360 across systems.</strong> <a href="https://cloud.google.com/blog/products/data-analytics/bigquery-graph-connecting-data-and-ai-at-scale">BigQuery Graph</a> defines a graph over existing tables, even Iceberg tables in other clouds, and traverses them in place. <a href="https://community.fabric.microsoft.com/t5/IQ-Community-Blog/Fabric-IQ-and-Ontology-when-your-data-speaks-the-language-of/ba-p/5176825">Fabric IQ’s ontology</a> binds each entity to its data and routes each query to the right engine. Netflix stores its metadata as RDF and serves everything else as a virtual graph (<a href="https://juansequeda.substack.com/p/trip-report-connected-data-london-2025">write-up</a>).</li><li><strong>Lineage and catalogue.</strong> Juan Sequeda’s advice: your first knowledge graph should be of your metadata (<a href="https://juansequeda.substack.com/p/trip-report-connected-data-london-2025">write-up</a>). Google traces what agents did with an <a href="https://cloud.google.com/blog/products/data-analytics/bigquery-graph-connecting-data-and-ai-at-scale">agent context graph</a> in BigQuery Graph.</li><li><strong>Fraud, ownership chains and supply chain risk.</strong> <a href="https://community.fabric.microsoft.com/t5/Fabric-Updates-Blog/Fabric-IQ-The-semantic-layer-powering-trusted-AI-agents-at/ba-p/5190739">Graph in Fabric</a> is generally available for these workloads. Google pairs <a href="https://docs.cloud.google.com/spanner/docs/graph/overview">Spanner Graph</a> for operations with <a href="https://docs.cloud.google.com/bigquery/docs/graph-overview">BigQuery Graph</a> for fraud, AML and supply chain analysis. Wiz runs its security graph on Amazon Neptune (<a href="https://juansequeda.substack.com/p/trip-report-connected-data-london-2025">write-up</a>).</li><li><strong>Contracts, regulations and research documents.</strong> AWS’s <a href="https://github.com/aws/context-ontology-accelerator">Context Ontology Accelerator</a> grounds new concepts in references like FIBO and marks which classes are backed by data. ABB built an LLM-generated graph for field service engineers (<a href="https://juansequeda.substack.com/p/trip-report-connected-data-london-2025">Juan Sequeda’s write-up</a>). ServiceNow’s Dean Allemang argues that LLM-drafted ontologies still need ontologist judgement (<a href="https://2026.connected-data.london/programme-per-day/">talk abstract</a>).</li><li><strong>Themes across many documents.</strong> <a href="https://aws.amazon.com/about-aws/whats-new/2025/03/amazon-bedrock-knowledge-bases-graphrag-generally-available">Bedrock Knowledge Bases GraphRAG</a> builds an entity graph from documents automatically in Neptune Analytics and combines it with vector search. <a href="https://docs.cloud.google.com/spanner/docs/graph/overview">Spanner Graph</a> combines graph and vector search for the same pattern.</li></ul><p>What the industry is telling us:</p><ul><li><strong>Most agent context doesn’t need a graph.</strong> Snowflake and Databricks ground their agents in a semantic layer for structured data plus search for documents. That’s the lower rungs of the ladder.</li><li><strong>Where graphs appear, they increasingly sit over the tables.</strong> BigQuery Graph, Fabric IQ’s bindings and ontology mappings all leave the data where it is. Virtual by default.</li><li><strong>Documents still mean materializing, and validation is still the hard part.</strong> Push-button GraphRAG builds the graph for you. The ontology-first tools add human review because extraction still needs it.</li><li><strong>Practitioners start small and grow.</strong> AstraZeneca, S&amp;P Global and Netflix started with governed vocabularies and metadata, and Zeiss recommends a high-value lighthouse use case that later ones can reuse (<a href="https://juansequeda.substack.com/p/trip-report-connected-data-london-2025">Juan Sequeda’s write-up</a>). AstraZeneca’s team has since taught <a href="https://2026.connected-data.london/venue/leonardo-royal-hotel-london-tower-bridge/">scaling from a project graph to an enterprise semantic layer</a>, and AEDO Group has described <a href="https://2026.connected-data.london/talks/experience-feedback-expanding-from-a-specific-knowledge-graph-towards-an-enterprise-wide-initiative/">small end-to-end graphs that each answer a business question</a>, joined up through inner-sourcing.</li></ul><h3>Grow it, don’t build it</h3><p>So, can you build an enterprise knowledge graph? Top-down, all at once: very unlikely. Grown one use case at a time on a thin shared core: yes. And for some use cases, the honest answer is to stop at a controlled vocabulary.</p><p>Gall’s Law puts it well. Complex systems that work evolve from simple systems that worked. Ones designed complex from scratch rarely work, and can’t be patched into working.</p><p>The honest pitch to leadership isn’t “we’re building the enterprise knowledge graph.” It’s “we’re building the first use case in a way that makes the second one cheaper.”</p><p>The architect’s question, “for what?”, eventually became the company’s strategy. It isn’t an easy journey. But with AI agents about to depend on what our data means, it’s one we need to get right.</p><p><em>If you’ve grown one, or decided you didn’t need one, I’d like to hear how it went. Tell me in the responses.</em></p><h3>Sources</h3><ul><li><a href="https://aws.amazon.com/about-aws/whats-new/2026/07/aws-context--ontology-accelarator-generally-available/">Amazon Web Services now introduces Context Ontology Accelerator</a>, AWS, 31 July 2026</li><li><a href="https://github.com/aws/context-ontology-accelerator">aws/context-ontology-accelerator</a>, GitHub</li><li><a href="https://github.com/aws/context-ontology-accelerator/blob/main/external-docs/content/ontologies.md">Context Ontology Accelerator: Ontologies documentation</a></li><li><a href="https://aws.amazon.com/about-aws/whats-new/2025/03/amazon-bedrock-knowledge-bases-graphrag-generally-available">Amazon Bedrock Knowledge Bases supports GraphRAG, now generally available</a>, AWS</li><li><a href="https://docs.snowflake.com/en/en/user-guide/snowflake-cortex/cortex-agents">Cortex Agents</a>, Snowflake documentation</li><li><a href="https://docs.snowflake.com/en/release-notes/2026/other/2026-08-28-cortex-analyst-transition-cortex-agents.md">Snowflake recommends transitioning from Cortex Analyst to Cortex Agents</a>, Snowflake, 28 August 2026</li><li><a href="https://upsolve.ai/blog/snowflake-semantic-layer">Snowflake Semantic Views: The 2026 Guide</a>, Upsolve</li><li><a href="https://www.databricks.com/blog/unity-catalog-metric-views-now-power-bi-one-unified-governed-semantic-layer-all-your-reporting">Unity Catalog Metric Views, now in Power BI</a>, Databricks</li><li><a href="https://answers.databricks.com/architecture-giving-ai-context-across-company-data-documents">Architecture for giving AI context across company data and documents</a>, Databricks</li><li><a href="https://atlan.com/know/ai-agent/databricks/databricks-data-ai-summit-2026-announcements/">Databricks Data + AI Summit 2026: Key Announcements</a>, Atlan</li><li><a href="https://www.devoteam.com/expert-view/microsoft-fabric-iq/">Microsoft Fabric IQ: when your semantic model becomes your agents’ source of truth</a>, Devoteam</li><li><a href="https://community.fabric.microsoft.com/t5/IQ-Community-Blog/Fabric-IQ-and-Ontology-when-your-data-speaks-the-language-of/ba-p/5176825">Fabric IQ and Ontology: when your data speaks the language of your business</a>, Microsoft Fabric Community</li><li><a href="https://community.fabric.microsoft.com/t5/Fabric-Updates-Blog/Fabric-IQ-The-semantic-layer-powering-trusted-AI-agents-at/ba-p/5190739">Fabric IQ: The semantic layer powering trusted AI agents at enterprise scale</a>, Microsoft Fabric Updates Blog</li><li><a href="https://cloud.google.com/blog/products/data-analytics/bigquery-graph-connecting-data-and-ai-at-scale">BigQuery Graph is now GA</a>, Google Cloud</li><li><a href="https://docs.cloud.google.com/bigquery/docs/graph-overview">Introduction to BigQuery Graph</a>, Google Cloud documentation</li><li><a href="https://docs.cloud.google.com/spanner/docs/graph/overview">Spanner Graph overview</a>, Google Cloud documentation</li><li><a href="https://juansequeda.substack.com/p/trip-report-connected-data-london-2025">Juan Sequeda, write-up of enterprise knowledge graph case studies</a>, 2 December 2025</li><li><a href="https://2026.connected-data.london/programme-per-day/">Dean Allemang (ServiceNow), Know-How Beats No-How: Steering Vibe Ontologies with Ontologist Judgment, talk abstract</a></li><li><a href="https://2026.connected-data.london/venue/leonardo-royal-hotel-london-tower-bridge/">AstraZeneca, How to cross the chasm: scaling from a project knowledge graph to an enterprise semantic data layer, masterclass listing</a></li><li><a href="https://2026.connected-data.london/talks/experience-feedback-expanding-from-a-specific-knowledge-graph-towards-an-enterprise-wide-initiative/">Experience Feedback: Expanding from a Specific Knowledge Graph towards an enterprise wide initiative</a>, Gaëtan Belbéoc’h, AEDO Group</li><li><a href="https://www.techmahindra.com/services/data-analytics/semara-ontology-knowledge-graph-accelerator/">SEMARA ontology and knowledge graph accelerator</a>, Tech Mahindra</li><li><a href="https://acuvate.com/accelerators/acucortex-ontology-accelerator/">AcuCortex Ontology Accelerator</a>, Acuvate</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=68c514251639" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>