<?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: Anton John Intes</title>
    <description>The latest articles on DEV Community by Anton John Intes (@apexsovereign).</description>
    <link>https://dev.to/apexsovereign</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%2F4141150%2Fb441a231-bd99-491b-b0bb-817a5dec47a0.png</url>
      <title>DEV Community: Anton John Intes</title>
      <link>https://dev.to/apexsovereign</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/apexsovereign"/>
    <language>en</language>
    <item>
      <title>Solving Spot GPU Eviction Context Loss with Kernel-Level eBPF Socket Hijacking</title>
      <dc:creator>Anton John Intes</dc:creator>
      <pubDate>Mon, 28 Sep 2026 18:02:59 +0000</pubDate>
      <link>https://dev.to/apexsovereign/solving-spot-gpu-eviction-context-loss-with-kernel-level-ebpf-socket-hijacking-1k32</link>
      <guid>https://dev.to/apexsovereign/solving-spot-gpu-eviction-context-loss-with-kernel-level-ebpf-socket-hijacking-1k32</guid>
      <description>&lt;h1&gt;
  
  
  Solving Spot GPU Eviction Context Loss with Kernel-Level eBPF Socket Hijacking
&lt;/h1&gt;

&lt;p&gt;Running LLM inference (vLLM/TGI) or model training on spot GPU instances offers huge cost savings, but a single provider reclamation signal can drop active TCP streams and wipe un-checkpointed sequence state.&lt;/p&gt;

&lt;p&gt;To eliminate this data loss, we built &lt;strong&gt;ApexSovereign&lt;/strong&gt;—a bare-metal GPU orchestration mesh engineered in Rust and Axum that handles spot failovers at the Linux kernel level.&lt;/p&gt;




&lt;h2&gt;
  
  
  Architecture Breakdown
&lt;/h2&gt;

&lt;h3&gt;
  
  
  1. eBPF Socket Hijacking (&lt;code&gt;BPF_MAP_TYPE_SOCKHASH&lt;/code&gt;)
&lt;/h3&gt;

&lt;p&gt;When a spot node receives an eviction signal from the data center provider, user-space proxies and DNS updates are too slow. &lt;/p&gt;

&lt;p&gt;By utilizing eBPF maps (&lt;code&gt;BPF_MAP_TYPE_SOCKHASH&lt;/code&gt;) and attaching &lt;code&gt;sk_skb&lt;/code&gt; stream parsers:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Active socket descriptors are updated directly in kernel space upon receiving an eviction signal.&lt;/li&gt;
&lt;li&gt;Reroute latency drops to &lt;strong&gt;1.67ms – 2.31ms&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;TCP connections remain open with &lt;strong&gt;zero client-side socket resets&lt;/strong&gt;.&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>ai</category>
      <category>rust</category>
      <category>ebpf</category>
      <category>infrastructure</category>
    </item>
    <item>
      <title>Building an Autonomous Compute Brokerage Engine with FastAPI and Supabase</title>
      <dc:creator>Anton John Intes</dc:creator>
      <pubDate>Thu, 24 Sep 2026 11:39:11 +0000</pubDate>
      <link>https://dev.to/apexsovereign/building-an-autonomous-compute-brokerage-engine-with-fastapi-and-supabase-3mak</link>
      <guid>https://dev.to/apexsovereign/building-an-autonomous-compute-brokerage-engine-with-fastapi-and-supabase-3mak</guid>
      <description>&lt;p&gt;Managing multi-provider compute allocations manually introduces unnecessary operational friction when scaling AI workloads, LLM inference, and automated backend execution. &lt;/p&gt;

&lt;p&gt;To streamline this process, I built and brought live &lt;strong&gt;&lt;a href="https://www.apexsovereign.ai" rel="noopener noreferrer"&gt;ApexSovereign.ai&lt;/a&gt;&lt;/strong&gt;—an autonomous compute brokerage engine designed to unify compute metering, double-entry financial ledgering, and sub-second API routing under a single system.&lt;/p&gt;




&lt;h3&gt;
  
  
  Technical Architecture &amp;amp; Stack
&lt;/h3&gt;

&lt;p&gt;Here is how the platform is architected to operate with zero fixed infrastructure overhead:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;FastAPI (Python 3):&lt;/strong&gt; Serves as our primary backend web service deployed on Render. Route handlers run via modularized startup entrypoints to maintain strict uptime execution and prevent module loading errors.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Supabase PostgreSQL:&lt;/strong&gt; Handles double-entry financial accounting via custom &lt;code&gt;ledger_entries&lt;/code&gt; tables alongside structured system audit logs (&lt;code&gt;audit_events&lt;/code&gt;).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Live Financial Settlement:&lt;/strong&gt; Integrated direct PayPal Business settlement rails, allowing real-time compute credit purchasing and instant ledger logging.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Edge Routing:&lt;/strong&gt; Vercel frontend proxy handling client-side requests and proxying traffic seamlessly to our backend engine.&lt;/li&gt;
&lt;/ol&gt;




&lt;h3&gt;
  
  
  Public Health &amp;amp; Diagnostic Verification
&lt;/h3&gt;

&lt;p&gt;To ensure platform availability, we exposed a public health matrix endpoint that validates active database bindings and gateway connections on every poll:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Live Platform:&lt;/strong&gt; &lt;a href="https://www.apexsovereign.ai" rel="noopener noreferrer"&gt;https://www.apexsovereign.ai&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Public Health Matrix:&lt;/strong&gt; &lt;a href="https://www.apexsovereign.ai/v1/platform/health-matrix" rel="noopener noreferrer"&gt;https://www.apexsovereign.ai/v1/platform/health-matrix&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;




&lt;h3&gt;
  
  
  Pilot Application Program
&lt;/h3&gt;

&lt;p&gt;Our &lt;strong&gt;Pilot Gateway&lt;/strong&gt; is officially open for teams requiring automated compute brokerage or specialized API routing. Submit an application directly through our site to get provisioned keys!&lt;/p&gt;

&lt;p&gt;I'd love to hear feedback from the DEV community on our double-entry ledger schema, FastAPI route design, and routing setup.&lt;/p&gt;

</description>
    </item>
  </channel>
</rss>
