<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Pontus Lindh</title>
    <description>The latest articles on DEV Community by Pontus Lindh (@nullfeather).</description>
    <link>https://dev.to/nullfeather</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3824479%2F5434a256-1960-4d34-9d6f-350a6334570f.png</url>
      <title>DEV Community: Pontus Lindh</title>
      <link>https://dev.to/nullfeather</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/nullfeather"/>
    <language>en</language>
    <item>
      <title>Show DEV: I built an evidence-first Windows diagnostic workstation</title>
      <dc:creator>Pontus Lindh</dc:creator>
      <pubDate>Sat, 14 Nov 2026 09:19:22 +0000</pubDate>
      <link>https://dev.to/nullfeather/show-dev-i-built-an-evidence-first-windows-diagnostic-workstation-51o0</link>
      <guid>https://dev.to/nullfeather/show-dev-i-built-an-evidence-first-windows-diagnostic-workstation-51o0</guid>
      <description>&lt;p&gt;I'm an onsite IT technician and independent developer from Sweden.&lt;/p&gt;

&lt;p&gt;I built FieldRook because I wanted a better way to investigate Windows problems without relying entirely on disconnected diagnostics, guesswork, or AI-generated answers that sound convincing but aren't supported by evidence.&lt;/p&gt;

&lt;p&gt;FieldRook is a local-first Windows desktop application built with WPF/WebView2, FastAPI, React, and TypeScript.&lt;/p&gt;

&lt;p&gt;It brings together structured investigations, bounded diagnostic tools, traceable evidence, optional AI reasoning, and explicit technician control.&lt;/p&gt;

&lt;p&gt;The current published pre-release is v0.6.0-rc.1, including Evidence Lens and Network Lab.&lt;/p&gt;

&lt;p&gt;The application is still early. Builds are unsigned, and broader real-world compatibility testing remains an important next step.&lt;/p&gt;

&lt;p&gt;I'm particularly interested in feedback from Windows technicians and developers:&lt;/p&gt;

&lt;p&gt;Would you trust the evidence and investigation workflow?&lt;/p&gt;

&lt;p&gt;What's missing compared with your real troubleshooting process?&lt;/p&gt;

&lt;p&gt;Which parts feel genuinely useful, and which don't?&lt;/p&gt;

&lt;p&gt;Repository: &lt;a href="https://github.com/Rookweld/FieldRook" rel="noopener noreferrer"&gt;https://github.com/Rookweld/FieldRook&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Latest pre-release: &lt;a href="https://github.com/Rookweld/FieldRook/releases/tag/v0.6.0-rc.1" rel="noopener noreferrer"&gt;https://github.com/Rookweld/FieldRook/releases/tag/v0.6.0-rc.1&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Apache-2.0 licensed. No signup required.&lt;/p&gt;

&lt;p&gt;I'd much rather get five detailed critiques than a hundred meaningless compliments.&lt;/p&gt;

</description>
      <category>showdev</category>
      <category>python</category>
      <category>opensource</category>
      <category>ai</category>
    </item>
    <item>
      <title>Why I stopped routing AI agents by keyword — and what I built instead</title>
      <dc:creator>Pontus Lindh</dc:creator>
      <pubDate>Sat, 14 Mar 2026 20:40:51 +0000</pubDate>
      <link>https://dev.to/nullfeather/why-i-stopped-routing-ai-agents-by-keyword-and-what-i-built-instead-2c5k</link>
      <guid>https://dev.to/nullfeather/why-i-stopped-routing-ai-agents-by-keyword-and-what-i-built-instead-2c5k</guid>
      <description>&lt;p&gt;I spent weeks building a local AI assistant before I admitted my routing system was broken.&lt;br&gt;
The setup was simple: if the user's message contained "code" or "debug", route to the dev agent. If it contained "write" or "edit", route to the editor. If it contained "plan" or "decide", route to the strategist.&lt;br&gt;
It worked fine in demos. It fell apart in real use.&lt;br&gt;
The problem with keyword routing&lt;br&gt;
Natural language doesn't map cleanly to intent. "Can you help me write a function?" triggers the editor instead of the dev agent because "write" appears before "function." "Should I design the API this way?" goes to the designer when it should go to the strategist. "Am I thinking about this wrong?" matches nothing at all.&lt;br&gt;
After enough of these failures I measured it: keyword routing was hitting around 60% accuracy on real conversation data. Tolerable on paper. Maddening to live with.&lt;br&gt;
The fix: LLM classification with a confidence threshold&lt;br&gt;
The approach that actually works is using a small, fast model as a classifier. You give it a system prompt listing your available agents, it reads the message, and it outputs a routing decision with a confidence score.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="n"&gt;pythonCLASSIFIER_SYSTEM&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="sh"&gt;"""&lt;/span&gt;&lt;span class="s"&gt;You are a routing engine for a multi-agent AI assistant.
Your only job: read a user message and output a JSON routing decision.

Available agents: architect, strategist, editor, researcher, devil, scheduler

Respond ONLY with valid JSON:
{&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;agent_id&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;: &lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;&amp;lt;id&amp;gt;&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;, &lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;confidence&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;: &amp;lt;0.0-1.0&amp;gt;, &lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;reason&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;: &lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;&amp;lt;one sentence&amp;gt;&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;}&lt;/span&gt;&lt;span class="sh"&gt;"""&lt;/span&gt;

&lt;span class="k"&gt;async&lt;/span&gt; &lt;span class="k"&gt;def&lt;/span&gt; &lt;span class="nf"&gt;classify&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;message&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="nb"&gt;str&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;-&amp;gt;&lt;/span&gt; &lt;span class="nb"&gt;tuple&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="nb"&gt;str&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nb"&gt;float&lt;/span&gt;&lt;span class="p"&gt;]:&lt;/span&gt;
    &lt;span class="n"&gt;resp&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="n"&gt;ollama&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;chat&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;
        &lt;span class="n"&gt;model&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;llama3.1:8b&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;  &lt;span class="c1"&gt;# Fast model for routing only
&lt;/span&gt;        &lt;span class="n"&gt;messages&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;
            &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;role&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;system&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;content&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="n"&gt;CLASSIFIER_SYSTEM&lt;/span&gt;&lt;span class="p"&gt;},&lt;/span&gt;
            &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;role&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;user&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;content&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="n"&gt;message&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;
        &lt;span class="p"&gt;],&lt;/span&gt;
        &lt;span class="n"&gt;options&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;temperature&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mf"&gt;0.1&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;num_predict&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;80&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;
    &lt;span class="p"&gt;)&lt;/span&gt;
    &lt;span class="n"&gt;parsed&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;json&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;loads&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;resp&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;message&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;][&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;content&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;])&lt;/span&gt;
    &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="n"&gt;parsed&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;agent_id&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;],&lt;/span&gt; &lt;span class="nf"&gt;float&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;parsed&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;confidence&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;])&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;But the classifier alone isn't enough. The key insight is: only switch agents when confidence clears a threshold.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="n"&gt;pythonCONFIDENCE_THRESHOLD&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="mf"&gt;0.72&lt;/span&gt;

&lt;span class="k"&gt;async&lt;/span&gt; &lt;span class="k"&gt;def&lt;/span&gt; &lt;span class="nf"&gt;route&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;message&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="nb"&gt;str&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;session&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;-&amp;gt;&lt;/span&gt; &lt;span class="nb"&gt;str&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="n"&gt;agent_id&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;confidence&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="nf"&gt;classify&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;message&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;

    &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;confidence&lt;/span&gt; &lt;span class="o"&gt;&amp;gt;=&lt;/span&gt; &lt;span class="n"&gt;CONFIDENCE_THRESHOLD&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
        &lt;span class="n"&gt;session&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;set_agent&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;agent_id&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;  &lt;span class="c1"&gt;# Switch
&lt;/span&gt;        &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="n"&gt;agent_id&lt;/span&gt;
    &lt;span class="k"&gt;else&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
        &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="n"&gt;session&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;current_agent&lt;/span&gt;  &lt;span class="c1"&gt;# Stay put, optionally suggest
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Below 0.72, you keep the current agent. The user might get a suggestion ("Looks like this might be a Strategist question — type /switch strategist to change"), but you don't force a jarring mid-conversation switch.&lt;br&gt;
Why 0.72 specifically?&lt;br&gt;
Through iteration: below 0.65 you get too many false switches. Above 0.80 you get too much stickiness where the agent won't change even when it clearly should. With more agents, consider going slightly lower. With fewer, slightly higher.&lt;br&gt;
One more wrinkle: after 3+ consecutive messages to the same agent, raise the threshold by 0.15. A conversation that's been in "coding mode" for a while needs a stronger signal to switch than a fresh message.&lt;br&gt;
The result&lt;br&gt;
Routing accuracy went from ~60% to consistently above 90% in real use. More importantly, it stopped feeling broken. The system now handles "what's wrong with my code AND should I restructure this whole thing?" correctly — the classifier sees the architectural question as the dominant intent and routes to the strategist, not the dev agent.&lt;br&gt;
The full pattern — including multi-agent pipelines, agentic tool loops, and the 3-layer memory architecture built on top of this router — is documented in a guide I put together at nullfeather.gumroad.com.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>python</category>
      <category>llm</category>
      <category>opensource</category>
    </item>
  </channel>
</rss>
