@AntonRGordon

Optimizing CUDA and RAG Systems with Profiling and Evaluation Frameworks

Nov 24, 2025 · 3 min read · By Anton R Gordon Understanding both hardware efficiency and answer quality is essential for building high-performance, trustworthy AI systems. CUDA workloads rely heavily on GPU utilization and kernel design, while retrieval-augmented generation (RA...

Scaling LLM Inference: Optimization Frameworks from First Principles

Oct 26, 2025 · 3 min read · Training large language models is expensive. Deploying them efficiently is even harder. As models like Llama 3, Mixtral, or Nemotron grow in size, the ability to optimize inference and training workflows determines whether they can be realistically u...

Anton R Gordon on Real-Time AI in Financial Services: High-Throughput Systems for Yield Curve Inversion and FX Hedging

Oct 18, 2025 · 3 min read · Real-time decisioning in financial services—especially for macro events like yield-curve inversions and fast FX exposures—demands an architecture that combines ultra-low latency market data ingest, high-throughput model inference, robust risk control...

Anton R Gordon’s Framework for Serverless AI Architectures with AWS Lambda

Oct 12, 2025 · 3 min read · In the evolving world of artificial intelligence and cloud computing, Anton R Gordon, an accomplished AI Architect and Cloud Specialist, continues to pioneer frameworks that make AI deployment faster, scalable, and cost-efficient. One of his most inf...

Why Self-Host Agentic RAG

Sep 17, 2025 · 3 min read · Agent teams (planner, retriever, ranker, synthesizer) thrive when latency is predictable, privacy is strict, and costs are controlled. That’s why I deploy them on Amazon EKS with vLLM as the serving layer: you keep data in-VPC, pin workloads to GPUs,...