<?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 Rukshan Dias on Medium]]></title>
        <description><![CDATA[Stories by Rukshan Dias on Medium]]></description>
        <link>https://medium.com/@rukshandias?source=rss-497445a20aa3------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*akfCjF0B8xHm4ybVuYhclA.jpeg</url>
            <title>Stories by Rukshan Dias on Medium</title>
            <link>https://medium.com/@rukshandias?source=rss-497445a20aa3------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 05:43:06 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@rukshandias/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[A Multi-Knowledge System Notebook on OKF: Map Them Like You Actually Learn]]></title>
            <link>https://medium.com/ifs-tech/a-multi-knowledge-system-notebook-on-okf-map-them-like-you-actually-learn-15b449898d83?source=rss-497445a20aa3------2</link>
            <guid isPermaLink="false">https://medium.com/p/15b449898d83</guid>
            <category><![CDATA[knowledge-retrieval]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[okf]]></category>
            <category><![CDATA[open-knowledge-format]]></category>
            <category><![CDATA[llm-wiki]]></category>
            <dc:creator><![CDATA[Rukshan Dias]]></dc:creator>
            <pubDate>Wed, 19 Aug 2026 02:01:05 GMT</pubDate>
            <atom:updated>2026-08-19T02:01:05.701Z</atom:updated>
            <content:encoded><![CDATA[<p>Most personal knowledge tools assume one vault, one graph, one brain. Humans don’t learn that way, we learn new knowledge by connecting it with knowledge we already know, as the brain thinks in relationships. We built a Multi-Knowledge System(Multi-KS) Notebook on top of <a href="https://cloud.google.com/blog/products/data-analytics/how-the-open-knowledge-format-can-improve-data-sharing">Open Knowledge Format (OKF)</a> introduced by Google, so each Knowledge System(KS) stays independent to clone and share, while Cross-KS associations live in a notebook-local layer that agents use for associative transfer when answering.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*siQVcteTh8UyZsyRTBHOLQ.png" /></figure><h3>Why one Knowledge System is not enough</h3><p>A single Knowledge System(KS) works until your knowledge stops being one domain.</p><p>Work docs, personal notes, a teammate’s shared corpus, a cloned research bundle each has its own owner, lifecycle, and right to stay publishable on its own. The tempting fix is to force them under one merged system. That creates two failures that compound over time.</p><p>First, <strong>you cannot cleanly detach</strong>. Once several KSs are flattened into a single corpus, removing a domain leaves broken links, orphaned references, and a graph that no longer reflects reality. Second, <strong>context bloats</strong>. Agents and readers keep pulling information from domains that should no longer be in scope, because the merged system has no safe way to remove a KS without damaging what remains.</p><p>What we needed was composition, not consolidation. Each KS stays a normal, cloneable OKF bundle. You clone it, onboard it, and map it to KSs you already know. Remove it later and notebook-local Cross-KS associations clean up with it — remaining markdown is untouched, because the cloned source never held those maps.</p><p>That is the problem the Multi-KS Notebook solves: keep Knowledge Systems independent to maintain, contribute, and share, and still transfer from what you already know into what you just cloned.</p><h3>What this actually buys you</h3><ul><li><strong>Faster onboarding:</strong> People switch projects often, and every switch costs weeks of “understanding the project.” Instead, a new joiner clones the KS the early team built, nodes and edges included - and an LLM can explain it like a book.</li><li><strong>Transfer learning, not just documents:</strong> This is where the idea departs from Confluence-style knowledge. Confluence pages don’t know what the reader already knows (more like a graph with no edges). Here this system enables to learn new thing based on what you already know (transfer learning concept): Like explaining gRPC concept(new knowledge) based on how REST(existing knowledge) works.</li><li><strong>Lower context bloat.</strong> You can offboard an obsolete KS as cleanly as you onboarded it. Which is impossible in single-KS system. Which helps to reduce hallucination without harming what remains.</li><li><strong>Continuous learning.</strong> Vendor neutrality leaves room for more flexibility: say, a personal continual-learning KS fed by a Cursor/Claude hook that captures new knowledge from your chats or /grill-me sessions.</li></ul><h3>Philosophy: associate, then transfer</h3><p>The philosophy behind the Multi-KS Notebook is simple. When a person already knows a KS from experience, they do not learn the next one from a blank slate. They start by connecting the new material to what they already understand(associate first) then they carry patterns across (transfer second).</p><p>A clean notebook example: your Finance KS already holds payment confirmation, credit hold, and refund habits. You clone a Supply Chain KS, which contains inventory reservation, fulfillment holds, release-to-ship rules. You do not start blank; you ask how their concepts map onto yours: reserved stock onto credit hold, release-to-ship onto payment cleared. That mapping is associative learning. Later, answering <em>“payment confirms late but inventory is already reserved, what should we do?”</em> by reading Supply Chain and following approved maps into Finance is transfer learning.</p><p>The Multi–KS Notebook makes that durable for AI-assisted reading: keep each KS intact, and record the learner’s Cross–KS maps where the agent can follow them when answering.</p><h3>What we kept short from OKF</h3><p>We build on <a href="https://cloud.google.com/blog/products/data-analytics/how-the-open-knowledge-format-can-improve-data-sharing">Open Knowledge Format (OKF)</a> — concept markdown, progressive disclosure throughindex.md, and agent-friendly traversal inside a single bundle. A separate study already covers that baseline; we will not retell it here.</p><p>What matters for this article is the enhancement: one OKF KS is still the unit of authorship and sharing. The Multi-KS Notebook adds how several of those units coexist, how Cross-KS associations are stored without rewriting source markdown, and how an agent uses those associations for transfer when answering.</p><figure><img alt="High-level architecture of Multi-Knowledge System Notebook" src="https://cdn-images-1.medium.com/max/1024/1*YVxCLsVS_QoCXGQqfen8CA.png" /><figcaption>High-level architecture of Multi-Knowledge System Notebook</figcaption></figure><h3>Independent Knowledge Systems, same rules</h3><p>A Multi-KS Notebook is a workspace that lists several KSs in config. Each entry is a normal OKF bundle: its own concepts, its own index.md, and if it uses a remote its own origin for sync or collaboration.</p><p>Every KS follows the same rules. Remotes are how a Knowledge System travels; they do not change what may be written into its markdown.</p><p>If a user wants to contribute to a KS, they follow those same rules inside that KS’s markdown, then push to its origin. Contributors push only the cloned content concept .md and index .md files. Association maps live outside the KS in the notebook-local layer, so they are not part of what gets pushed.</p><h3>The Separation rule</h3><p>This is the architectural hinge and where we go deep on purpose.</p><ul><li><strong>Intra-Knowledge System edges</strong> live in concept markdown as ordinary links. They ship with that KS when you push or share it.</li><li><strong>Cross-Knowledge System associations</strong> never appear as paths in source .md files. They live only in the notebook’s Association store, inside the notebook-local layer.</li></ul><p>That single rule keeps a cloned KS honest. Someone else can pull the same markdown without inheriting <em>your</em> personal maps onto <em>their</em> other KSs. It also makes detach safe: notebook maps can be removed with the KS, while remaining KSs keep intact Intra-KS links and untampered source files.</p><h3>Clone, onboard, and map without rewriting the source</h3><figure><img alt="Cloning, onboarding a new KS and creating relationshipd between KSs" src="https://cdn-images-1.medium.com/max/1024/1*SAM--0m2ktgbq0fgldwwXQ.png" /><figcaption>Cloning, onboarding a new KS and creating relationships between KSs</figcaption></figure><p>Cloning is the on-ramp. You bring a KS into the notebook as markdown and index.md only. No notebook associations travel with the clone; those are always notebook-local.</p><p>Onboard then answers one question: <em>how does this new Knowledge System connect to what we already know?</em> We do not auto-merge corpora, and we do not ask a human to review every file. The flow is:</p><p><strong>1. Register</strong> the KS in notebook config.</p><pre>&quot;ks&quot;: {<br>    &quot;knowledge_system_01&quot;: {<br>      &quot;label&quot;: &quot;knowledge_system_01&quot;,<br>      &quot;path&quot;: // path where cloned/created KS exists<br>    },<br>    &quot;knowledge_system_02&quot;: {<br>      &quot;label&quot;: &quot;knowledge_system_02&quot;,<br>      &quot;path&quot;: // path where cloned/created KS exists<br>    }<br>  }</pre><p><strong>2. Catalog-first propose</strong> — rank candidate Cross-KS pairs from index.md trees and concept frontmatter (title, description, tags, type). Similarity is a token-overlap ranker that would compare title, description, tags between 2 concepts, not proof of a peer.</p><pre>&quot;candidates&quot;: [<br>    {<br>      &quot;from_ks&quot;: &quot;knowledge_system_01&quot;,<br>      &quot;from_path&quot;: &quot;mcp-architecture-explained.md&quot;,<br>      &quot;to_ks&quot;: &quot;knowledge_system_02&quot;,<br>      &quot;to_path&quot;: &quot;open-telemetry-observability-architecture.md&quot;,<br>      &quot;similarity_score&quot;: 55<br>    },<br>    ....<br>    {<br>      &quot;from_ks&quot;: &quot;knowledge_system_01&quot;,<br>      &quot;from_path&quot;: &quot;mcp-architecture-explained.md&quot;,<br>      &quot;to_ks&quot;: &quot;knowledge_system_02&quot;,<br>      &quot;to_path&quot;: &quot;distributed-tracing-architecture.md&quot;,<br>      &quot;similarity_score&quot;: 45<br>    }<br>]</pre><p><strong>3. Agent shortlist</strong> — an agent judges which candidates are semantically valid transfer peers (peek at bodies only when the catalog is ambiguous). Weak links are dropped, not padded.</p><p><strong>4. Human approve</strong> — accept selected associations into the Association store, or discard. The shortlist being empty is also valid.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ne06jlT5WMIYNsMux01Vjg.png" /></figure><p>Accepted maps never rewrite the cloned source. They land only in the notebook-local layer. The KS stays clean to contribute to, push, or re-share. If you later remove it from the notebook, associations that pointed at it can clean up with it without leaving broken Cross-KS links inside other KSs’ markdown.</p><h3>How answers navigate across Knowledge Systems</h3><p>Answering is where associate and transfer show up in the agent loop. Progressive disclosure inside a KS still comes first. Following Cross-KS associations is a deliberate second step, not a dump of every corpus into context.</p><p>Preferred order:</p><ol><li><strong>Orient</strong> inside a relevant KS via its index.md (and nested indexes when needed).</li><li><strong>Go deeper</strong> with Intra-KS links in concept bodies.</li><li><strong>Search</strong> when you already know a term or need to span several KSs by keyword, for example with grep-style search, only when needed.</li><li><strong>Transfer</strong> via the Association store when the question needs a peer concept in another KS. Again only when needed.</li><li><strong>Cite</strong> real concept markdown. The derived index helps you find; it is not what you quote as truth.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2uE-drSDAWo7oxfltVekXA.png" /><figcaption>How Multi-KS notebook system answer a question</figcaption></figure><p>That is transfer learning in the notebook: the agent does not flatten Supply Chain and Finance into one file. It reads the reserved-stock concept, follows an approved association to the credit-hold pattern you already trust, and synthesizes an answer that uses both.</p><p><strong>Ingesting knowledge</strong> follows the same Separation discipline. New Intra-KS “See also” links may land in markdown; Cross-KS pairs queue into the Association store instead. Maintenance stays multi-KS: each bundle remains independently editable, contributable, and shareable while the notebook accumulates learner-specific bridges over time — and remains able to detach a KS without corrupting the rest.</p><h3>Takeaways</h3><p>Once we used the notebook for real questions, a few rules stuck:</p><ul><li><strong>Compose, don’t merge</strong> — each OKF KS stays an independent, cloneable, shareable unit. Merge them into one system and you lose clean detachment while context keeps bloating.</li><li><strong>Clone, onboard, map without tampering</strong> — associations live in a notebook-local layer, cloned source stays clean. Remove a KS and notebook maps clean up without broken links in remaining markdown.</li><li><strong>Contribute with the same rules</strong> — edit Intra-KS content and push to that KS’s origin; association maps stay outside and are not pushed.</li><li><strong>Propose, shortlist, approve</strong> — catalog + agent can suggest transfer peers that humans could accept. Prefer zero wrong bridges over many automatic ones.</li><li><strong>Orient, then transfer</strong> — navigate inside a KS first, follow associations when needed, and cite real concept markdown.</li></ul><p>Map them like you actually learn, and keep the KSs independent enough that tomorrow’s clone still fits into today’s notebook.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=15b449898d83" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ifs-tech/a-multi-knowledge-system-notebook-on-okf-map-them-like-you-actually-learn-15b449898d83">A Multi-Knowledge System Notebook on OKF: Map Them Like You Actually Learn</a> was originally published in <a href="https://medium.com/ifs-tech">IFS Tech Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>