Add edge AI retraining section to R&D page

Author: Reg (Ceradon AI)

technology.html

@@ -82,6 +82,41 @@ <h3 class="text-xl font-semibold text-[color:var(--white)]">Localized Production
               <h3 class="text-xl font-semibold text-[color:var(--white)]">Passive Detection Expansion</h3>
               <p class="text-sm text-subtle">Wi-Fi CSI anomaly detection for zero-emission awareness of Group 1&ndash;3 threats.</p>
             </div>
+            <div class="timeline-item">
+              <h3 class="text-xl font-semibold text-[color:var(--white)]">Operator-Directed Edge Retraining</h3>
+              <p class="text-sm text-subtle">On-device model adaptation in minutes &mdash; no reach-back to CONUS, no contractor, no weeks-long retraining pipeline.</p>
+            </div>
+          </div>
+        </div>
+
+        <!-- Edge AI Retraining -->
+        <div class="card-glass p-8 space-y-6">
+          <div class="space-y-2">
+            <span class="tag">Emerging Capability</span>
+            <h2 class="text-2xl font-semibold text-[color:var(--white)]">Edge AI Retraining &mdash; Closing the Last Mile</h2>
+          </div>
+          <p class="text-subtle max-w-3xl">Every defense AI program hits the same wall: models trained in controlled environments degrade in the field. Different terrain, lighting, concealment methods, and threat profiles mean laboratory accuracy rarely survives first contact with reality.</p>
+          <p class="text-subtle max-w-3xl">The current DoD solution &mdash; ship data back to CONUS, retrain in a SCIF, push updates weeks later &mdash; doesn&rsquo;t work for small teams operating in denied or disconnected environments. We&rsquo;re building systems that retrain on-device, in minutes, directed by the operator who knows the environment best.</p>
+
+          <div class="grid gap-6 md:grid-cols-2 mt-4">
+            <div class="space-y-3">
+              <h3 class="text-lg font-semibold text-[color:var(--white)]">Why This Matters</h3>
+              <ul class="space-y-2 text-sm text-subtle">
+                <li class="flex gap-3"><span class="text-[color:var(--ceradon-blue)]">&#9657;</span> SOF teams need systems that adapt to the environment they&rsquo;re in <em>right now</em> &mdash; no reach-back to data centers.</li>
+                <li class="flex gap-3"><span class="text-[color:var(--ceradon-blue)]">&#9657;</span> Border and perimeter security faces constantly changing threat profiles that demand rapid model updates.</li>
+                <li class="flex gap-3"><span class="text-[color:var(--ceradon-blue)]">&#9657;</span> Partner forces can train on <em>their</em> environment with <em>their</em> data &mdash; no classified training sets required. A significant FMS enabler.</li>
+                <li class="flex gap-3"><span class="text-[color:var(--ceradon-blue)]">&#9657;</span> Persistent surveillance at FOBs and COPs where the threat signature changes weekly.</li>
+              </ul>
+            </div>
+            <div class="space-y-3">
+              <h3 class="text-lg font-semibold text-[color:var(--white)]">Our Approach</h3>
+              <ul class="space-y-2 text-sm text-subtle">
+                <li class="flex gap-3"><span class="text-[color:var(--ceradon-blue)]">&#9657;</span> <strong>Operator-in-the-loop:</strong> Human oversight is the safety mechanism &mdash; operators validate retraining results before deployment.</li>
+                <li class="flex gap-3"><span class="text-[color:var(--ceradon-blue)]">&#9657;</span> <strong>Incremental learning:</strong> Adapt models with small, field-collected datasets without catastrophic forgetting of base capabilities.</li>
+                <li class="flex gap-3"><span class="text-[color:var(--ceradon-blue)]">&#9657;</span> <strong>Edge-native compute:</strong> Full retraining pipeline runs on embedded hardware &mdash; no cloud dependency, no exfiltration risk.</li>
+                <li class="flex gap-3"><span class="text-[color:var(--ceradon-blue)]">&#9657;</span> <strong>Closed-loop integration:</strong> Retrained models feed directly into autonomous response and C2 systems.</li>
+              </ul>
+            </div>
           </div>
         </div>