https://khomkrit.cc/en/case-studies/Recent content in Case-Studies on Khomkrit | Cloud & Infrastructure ArchitectHugoen-USTue, 12 May 2026 12:05:22 +0700https://khomkrit.cc/en/case-studies/vectormesh/Tue, 12 May 2026 12:05:22 +0700https://khomkrit.cc/en/case-studies/vectormesh/<p><img alt="Vector Mesh overview architecture" loading="lazy" src="https://khomkrit.cc/images/279eff97-0663-4c9d-aaac-352125cbc4c6.jpeg"></p> <h1 id="vectormesh--designing-a-high-traffic-secure-by-design-enterprise-ai-rag-platform">VectorMesh – Designing a High-Traffic, Secure-by-Design Enterprise AI-RAG Platform</h1> <h3 id="tldr">TL;DR</h3> <p>This article explores the architecture and engineering decisions behind <strong>VectorMesh</strong>, an enterprise-grade AI-RAG (Retrieval-Augmented Generation) platform designed to address the bottlenecks that emerge when AI systems move from experimentation into production.</p> <p>The platform is built around <strong>Rust-based microservices, Kubernetes, Istio, Apache APISIX, and end-to-end observability</strong>, with a strong focus on scalability, security, performance, and operational visibility.</p>https://khomkrit.cc/en/case-studies/cudf/Mon, 12 Feb 2024 09:32:04 +0700https://khomkrit.cc/en/case-studies/cudf/<p>Tags: Data Engineering, Python, NVIDIA RAPIDS, Performance Optimization</p> <p>The Problem: The Pandas Bottleneck When building data pipelines—whether for AI/RAG preprocessing, algorithmic trading, or standard ETL—Pandas is the undisputed king of tabular data. However, as your dataset scales to millions of rows, Pandas becomes a notorious CPU bottleneck.</p> <p>Waiting seconds (or minutes) for simple aggregations disrupts the development flow and inflates cloud compute costs. Recently, I explored how to bypass this limitation using NVIDIA RAPIDS cuDF, a library that mirrors the Pandas API but executes strictly on the GPU.</p>