Release v1.2.0: Newspaper-style layout with major UI refinements

This release transforms PING into a sophisticated newspaper-style digital
publication with enhanced readability and professional presentation.

Major Features:
- New FeaturedPostHero component with full-width newspaper design
- Completely redesigned homepage with responsive newspaper grid layout
- Enhanced PostCard component with refined typography and spacing
- Improved mobile-first responsive design (mobile → tablet → desktop → 2XL)
- Archive section with multi-column layout for deeper content discovery

Technical Improvements:
- Enhanced blog post validation and error handling in lib/blog.ts
- Better date handling and normalization for scheduled posts
- Improved Dockerfile with correct content volume mount paths
- Fixed port configuration (3025 throughout stack)
- Updated Tailwind config with refined typography and newspaper aesthetics
- Added getFeaturedPost() function for hero selection

UI/UX Enhancements:
- Professional newspaper-style borders and dividers
- Improved dark mode styling throughout
- Better content hierarchy and visual flow
- Enhanced author bylines and metadata presentation
- Refined color palette with newspaper sophistication

Documentation:
- Added DESIGN_BRIEF_NEWSPAPER_LAYOUT.md detailing design principles
- Added TESTING_RESULTS_25_POSTS.md with test scenarios

This release establishes PING as a premium publication platform for
AI orchestration and contextual intelligence thought leadership.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

content.bak/posts/2025/02/2025-02-26-on-prem-gpus.md

@@ -0,0 +1,54 @@
+---
+title: "Why On-prem GPUs Still Matter for AI"
+description: "Own the stack. Own your data."
+date: "2025-02-26"
+publishDate: "2025-02-28T09:00:00.000Z"
+author:
+  name: "Anthony Rawlins"
+  role: "CEO & Founder, CHORUS Services"
+tags:
+  - "gpu compute"
+  - "contextual-ai"
+  - "infrastructure"
+featured: false
+---
+
+Cloud GPUs are everywhere right now, but if you’ve tried to run serious workloads, you know the story: long queues, high costs, throttling, and vendor lock-in. Renting compute might be convenient for prototypes, but at scale it gets expensive and limiting.
+
+That’s why more teams are rethinking **on-premises GPU infrastructure**.
+
+## The Case for In-House Compute
+
+1. **Cost at Scale** – Training, fine-tuning, or heavy inference workloads rack up cloud costs quickly. Owning your own GPUs flips that equation over the long term.
+2. **Control & Customization** – You own the stack: drivers, runtimes, schedulers, cluster topology. No waiting on cloud providers.
+3. **Latency & Data Gravity** – Keeping data close to the GPUs removes bandwidth bottlenecks. If your data already lives in-house, shipping it to the cloud and back is wasteful.
+4. **Privacy & Compliance** – Your models and data stay under your governance. No shared tenancy, no external handling.
+
+## Not Just About Training Massive LLMs
+
+It’s easy to think of GPUs as “just for training giant foundation models.” But most teams today are leveraging GPUs for:
+
+* **Inference at scale** – low-latency deployments.
+* **Fine-tuning & adapters** – customizing smaller models.
+* **Vector search & embeddings** – powering RAG pipelines.
+* **Analytics & graph workloads** – accelerated by frameworks like RAPIDS.
+
+This is where recent research gets interesting. NVIDIA’s latest papers on **small models** show that capability doesn’t just scale with parameter count — it scales with *specialization and structure*. Instead of defaulting to giant black-box LLMs, we’re entering a world where **smaller, domain-tuned models** run faster, cheaper, and more predictably.
+
+And with the launch of the **Blackwell architecture**, the GPU landscape itself is changing. Blackwell isn’t just about raw FLOPs; it’s about efficiency, memory bandwidth, and supporting mixed workloads (training + inference + data processing) on the same platform. That’s exactly the kind of balance on-prem clusters can exploit.
+
+## Where This Ties Back to Chorus
+
+At Chorus, we think of GPUs not just as horsepower, but as the **substrate that makes distributed reasoning practical**. Hierarchical context and agent orchestration require low-latency, high-throughput compute — the kind that’s tough to guarantee in the cloud. On-prem clusters give us:
+
+* Predictable performance for multi-agent reasoning.
+* Dedicated acceleration for embeddings and vector ops.
+* A foundation for experimenting with **HRM-inspired** approaches that don’t just make models bigger, but make them smarter.
+
+## The Bottom Line
+
+The future isn’t cloud *versus* on-prem — it’s hybrid. Cloud for burst capacity, on-prem GPUs for sustained reasoning, privacy, and cost control. Owning your own stack is about **freedom**: the freedom to innovate at your pace, tune your models your way, and build intelligence on infrastructure you trust.
+
+The real question isn’t whether you *can* run AI on-prem.
+It’s whether you can afford *not to*.
+